E. A. Fisher, H. G. Thornton, and W. A. Mackenzie 341 



7. The exceptionally Variable Sets in Cutler's Data 

 The records of the exceptionally variable sets of plates which occurred 

 in Cutler's data, when identified by the method of the preceeding section, 

 were studied individually with a view to gaining hght upon the cause of 

 their occurrence. As it is not necessary to reproduce the whole of the 

 statistical tests which were applied, we shall confine ourselves to the 

 main facts which emerged, and which served to justify the previous 

 conclusions, as well as to indicate the nature of the disturbing cause. 

 The following facts appear to be unquestionable: — 



(1) The proportion of exceptionally variable sets is the same for the 

 sets of three, four and five plates in each portion of the total period. 



(2) The proportion of exceptionally variable sets varies greatly at 

 different periods, the exceptions occurring in well marked epidemics. 



The evidence for these statements may be put in the form of a triple 

 contingency table (see Fig. 2) 



Table XI 



in which the whole of the 362 observations are divided, 



(1) according to the number of plates observed, 



(2) in ten periods of time of alternately 36 and 37 days, into which 

 the year was divided, 



(3) according as they are judged to be exceptionally variable, or not, 

 solely upon the evidence of the x^ index. The subdivision which would be 

 made taking also into account the evidence for epidemics is shown in 

 brackets, but in discussing the evidence for epidemics these modifications 

 are ignored. 



