6 JVewcomb and Button — Speed of 



of five minutes. Thus we have 32 giving 9:50, but none giv- 

 ing 9:49, and only six giving 9:51. There are 13 giving 10 

 o'clock and there would have been many more of them if the 

 catalogue had included those which stated the time as being 

 " about " 10 o'clock or " near " 10 o'clock. There are 86, or 

 more than one fourth the whole number, which give 9:55. 

 Every one of the 9:50 reports is rejected. It is certain that 

 they all involve errors greater than one minute too early, and 

 the large number of them would introduce a large systematic 

 error into the mean ; and as there is no apparent reason for re- 

 jecting or keeping one observation rather than another, all of 

 them are thrown out. All of the 10 o'clock observations are 

 thrown out. For, upon further examination, all giving 9:58 and 

 seconds, 9:59, 10:01 and 10:02 will be rejected on their merits. 

 This would leave the 10 o'clock reports as an isolated group in 

 an otherwise comparatively orderly series, and its effect would 

 be to introduce an error of unknown magnitude and of anoma- 

 lous character. In dealing with those giving 9:55 there is 

 more difficulty. The following course has been adopted. 

 Wherever a report states clearly, or raises a strong presump- 

 tion, that this was really the nearest minute observed, to the 

 exclusion of any other, it is accepted if otherwise unobjection- 

 able. Where this evidence is wanting the report is rejected. 

 It is quite probable that some thus rejected are very good ob- 

 servations ; but it is clearly better to reject many possibly good 

 observations (provided a sufficient number remain) than to 

 admit a few bad ones with the certainty of introducing an un- 

 known error. The number of 9:55 reports thus rejected is 13, 

 which happens to be just one half. 



Still other observations are rejected on their merits. A ma- 

 jority of these are thrown out for what are presumed to be 

 large unexplained errors. There are 29 of them, of which 15 

 are rejected for being two minutes or more too early and 14 

 for being as much, or more than as much, too late, when com-' 

 pared with a larger number of much better observations in the 

 same locality or in the immediately surrounding region. The 

 rejection of these 29 observations does not greatly affect the 

 deduced speed, but.it does diminish notably the computed 

 probable error. The total number rejected for all causes is 

 130 and the number accepted is 186. These have been sepa- 

 rated into four groups, each containing data which are consid- 

 ered to be as nearly homogeneous as possible ; that is to say, in 

 each group the observations are presumed to have the same 

 sources of error, whether accidental or systematic. 



The first group is required to fulfill the following condi- 

 tions : (1) The report must specify the beginning, or the time 

 when the tremors first became sensible. (2) It must give not 



