R. A. Fisher, H. Cr. Thornton, and W. A. Mackenzie 353 



about 20 per cent. Series B also shows certain other irregularities and 

 possibly the occurrence of sporadic causes of variation. Series C, which 

 represents the final perfection of the technique employed, shows no 

 excessively variable sets of plates. 



Table XX 



It is, we believe, possible to indicate the cause of the small systematic 

 excess of variance in this exceptionally fine body of data. As has been 

 observed, the duplicate counts, which are recorded in full, do not agree 

 very closely, and it is possible that what may be called " error of counting " 

 is responsible for the existing discrepancy. If we consider such a typical 

 pair of duplicate counts such as that shown in Table XXI, we may regard 



Table XXI 



the mean difference, as due to the personal equation of the analyst; and 

 the departures from the mean as made up of the several "errors of 

 counting" of the set. If the standard "error of counting" is a, then the 

 mean value of the sum of the squares of the three departures will be ia^. 

 In this way the standard "error of counting" was estimated for each of 

 the main groups of observations in Series C, divided according to the 

 mean number of colonies per plate, and the additional variance ascrib- 

 able to "errors of counting" expressed as a percentage of the expected 

 variance. 



Ann. Biol, ix 



23 



