From Table XI, the probable errors of ;i single observation by the different 

 pendulums are calculated as before. 

 We find, thus, e b - 5-4xlO~ 7 sec. 



e, =1-6 



e tl = = 4-3 



em == 3-0 x 10-' sec. 



The probable error of the mean result by the different pendulums is also calculated, 

 and the values come out as follows : 



p- a -- -- 1-9X10- 7 sec. 



p. = (5 



Pa. ==1-5 



p m -.-. 1-OxlO- 7 sec. 



In this series, simple inspection is sufficient to show us that a systematic error 

 of some kind was present. It will be noted that the observations run very closely in 

 pairs. Thus, evening observations agree well with the following morning observations, 

 while there is disagreement between a.m. observations and the succeeding p.m. observa- 

 tions on the same day. 



To show this discrepancy better we will form the following Table : 



TABLE XII. Time of Swing. 



The only explanation possible for the sudden changes from a.m. observation to 

 the following p.m. observation is that changes have occurred which have not been 

 allowed for. These changes, moreover, must have occurred between the two times 

 of observation, or at about noon on each day. Temperature and barometer changes 

 we can at once neglect as insufficient to cause these differences, but there remain varia- 

 tions due to changes in clock rate and to changes in the apparatus itself, such as change 

 of level and of rigidity. It will be noticed that pendulums No. 5 and 21 seem to show 

 the errors most, while No. 7 appears to have remained little changed throughout. This 

 might argue a possible change in the position of the levelling screw nearest to pendulums 



21 B 3 



