subject, were (apart from errors in the time observations) caused by unevenness of 

 clock rate, and it was found that these irregularities were particularly serious imme- 

 diately after the clock was rewound, so that the first observations often had to be 

 discarded. 



Whatever may be the cause of the irregularities in clock rate, it is clear that they 

 may affect the numbers given in the Table above in different degree. Thus, if the 

 clock rate remains constant during the whole course of the observations (the clock being 

 rated only at the beginning and end of the observations), there is no reason why the 

 figures given in the last two columns should notably differ from one another ; if the 

 clock rate remains constant during the period involved in swinging all three pendulums 

 (one set), but has a different value from one set to another, the third column should 

 show higher values than the last. Significantly higher values are, in fact, shown in 

 the majority of cases, especially at Cape Evans and for Christchurch, 1910, where big- 

 differences exist between the numbers in these two columns. The reason can plausibly 

 be put down to this cause, and where such differences do not exist, it may be inferred 

 that the value in the last column gives a good measure of the accuracy of comparison 

 between pendulums and clock. 



The difference between p M and e m is, of course, entirely conditioned by the number 

 of sets observed. 



Judged by the values of e x alone, the most satisfactory observations were, in order, 

 those of Potsdam I, Potsdam II, Melbourne, Cape Evans C and Wellington. The 

 Cape Evans observations (series C) were, however, clearly vitiated by irregular clock 

 rate and, if this had not been the case, might have ranked still higher. 



The values of p m do not, however, represent the whole probable error of the final 

 result, for they do not adequately include probable errors connected with : 



(1) The determination of the flexure correction. 



(2) The determination of the temperature constants of the pendulums. 



(3) The determination of the density constants for the pendulums. 



(4) The application of the correction for temperature lag. 



(5) The determination of the clock rate. 



(6) The changes in the pendulums during the period between the initial and the 

 final observations at Potsdam. 



(7) The determination of the time of swing at Potsdam, the base station. 



Evaluation of the. Probable Errors. 



(1) Probable error of the flexure correction. This has been already worked out for 

 the different field stations. 



(2) and (3) The probable errors of the detenu 'million <>J temperature and barometer 

 constants are only of importance when the temperature and barometer readings at 

 the field stations are different from those recorded during the base station observations. 



85 * 3 



