We may therefore say that, in this series (Potsdam I), there were probably no 

 systematic differences in the working of the clock or in the method of comparing the 

 clock times with the time of vibration of the pendulum that is, the coincidence apparatus 

 was also working well. The importance of this last point can be readily estimated 

 when one considers the effect of a small retardation in the time elapsing between the 

 " break " of the clock circuit, and the movement of the lever in the coincidence apparatus. 

 Variations in the action of the attracting electromagnets, due to running down of 

 the battery, rise of temperature in the coils, together with change in the permeability 

 of the iron core, may all produce apparent fluctuations of considerable magnitude in 

 the clock rate. 



Thus, during the course of the observation on a single pendulum, if the action 

 of the lever has become retarded by 0-005 sec. during the 30 mins. interval between 

 the initial and final series of readings, the apparent daily rate will differ from the true 

 one by 0-24 sec., and this requires a correction to the time of swing of 14 units in the 

 7th decimal place. 



The observations at Potsdam show that the effect of all these factors is negligible 

 under such favourable conditions. There are, however, theoretical objections to this 

 form of coincidence apparatus, and the fact that serious discrepancies in the observations 

 made at Cape Evans pointed to a defect in the coincidence apparatus, led to the 

 evolution of a form in which no moving parts are involved. This arrangement will be 

 found described under the discussion of the observations at Christchurch, 1913. 



It must be pointed out that the foregoing calculations of the probable errors are 

 in no case a measure of the actual " errors of observation." To get an estimate of 

 the merit of the comparison between pendulum and clock, it is, of course, necessary 

 that all the variables rate, temperature, barometer, etc. should remain constant 

 throughout the series of observations. In practice, of course, this never occurs, but 

 one is able by a suitable handling of the observations to obtain a value which is a fair 

 measure of the probable error of the comparison between clock and pendulums. 

 This is most simply done by forming the differences between the time of swing 

 of the mean pendulum and Nos. 5, 7 and 21. These differences will be almost 

 independent of clock rate and such other variables, and the estimation of the probable 

 error of such a difference will be some measure of the observational accuracy 

 alone. 



In Table VI below these differences are put down, and the probable error of a 

 single difference evaluated in the usual manner by the help of Table VII. 



17 



