356 



Method of estimating Baeterial Density 



indices of variance on the separate media. In fact, however, no such 

 remarkably high variance was found. 



The analyses were performed with sets of six plates, and we have 

 chosen the first 100 of these sets for examination. The expected and 

 observed numbers are shown in Table XXV. 



Table XXV 



The excess of highly variable sets occasions no surprise; we have met 

 with this feature in about the same proportion in Cutler's data. What 

 is astonishing in this case is the immense excess of sets less variable, 

 and in the majority of cases much less variable, than would be the case 

 under undisturbed conditions of random sampling. 



In the fourth column we have shown the expected distribution fitted 

 to the total number in the range from 2 to 14. This seems to agree with 

 the distribution observed within this range. We are unwilling to lay 

 much stress on this explanation since the agreement is based on only 

 36 observations. If it were accepted it would imply that the conditions 

 which lead to the Poisson Series were really operative in about 44 per 

 cent, of the cases, that in at least 10 and probably 11 per cent, excessive 

 variability has been produced, and in the remaining 45 per cent, the 

 variability has been abnormally depressed. 



The extent to which the differences between the counts of parallel 

 plates is diminished seems to put the phenomenon beyond the reach of 

 the ordinary explanations; there are some indications, for example, that 

 the plates have not been in all cases completely counted, but it is 



