Introducing the uncertainty in the azimuth constant, the total probable error of 

 the daily rate is 



\ (0-07) 2 -f-(0-02) 2 



and 



v/(0-07) 2 +(0-01) 2 



v/(0-07) 2 -|-(0-07) 2 , for the three intervals. 



The maximum probable error is 1 sec., corresponding to 6 X 10~ 7 sec. in the time of swing. 

 As regards the observations at Cape Evans in 1911 (series A and B), it is quite 

 impossible to obtain any indication of the probable error in clock rate from the observa- 

 tions for time and, for this reason, these two series have been neglected. It is interesting, 

 however, to note that the final figures for series A are not very widely different from 

 those of series C and D (which agree within the limits of the calculated probable errors). 

 On the other hand, the more consistent results of series B diverge widely from 

 the remainder, and indicate a constant error in the applied rate which seems outside 

 the bounds of probability. Other causes may, however, have contributed to the 

 consistent errors which appear to be inherent in this series for instance, the unsatis- 

 factory working of the coincidence apparatus. 



(6) The probable error due to variability of the pendulums. 



From Table LXI, it is seen that only in the case of pendulum No. 7 was the time 

 of swing at Potsdam the same on both occasions. Nos. 21 and 5 had on the second 

 occasion, times of swing 16xlO~ 7 sec. and 11 x 10~ 7 sec. less, respectively, than on the 

 first occasion. 



The changes are not sufficiently definite to enable one to say exactly when they 

 occurred, but such changes are usually treated and evaluated as errors of an accidental 

 type, which change from station to station and are constant during all the observations 

 of the station. 



A value ^ is first calculated from the data in Tables III, XT, etc., giving the 

 squares of the differences of each observation by each pendulum from the mean of 

 the station for each pendulum. These squares, 2 (OT), are then summed for each station 

 and these sums, ^ (VV), are given in Table LXIX below : 



TABLE LXIX. Values of 2 (VV). 



88 



