352 Method of estimating Bacterial Density 



Such a departure from expectation would occur by chance but once 

 in 600 tests; it is therefore clearly significant. The technique used here 

 did not therefore give results of such accuracy that the variance between 

 parallel plates could approximate to the Poisson Series. 



( B ) The data of Engberding ( 12 ) 



The parallel platings given by this autlior were made to test various 

 points connected with the plate method of counting soil bacteria. Some 

 of the sets of platings were made on a variety of gelatine and agar media, 

 as a test of these. The majority, however, were poured on an agar medium, 

 containing "Nahrstoff-Heyden," that was considered by the author to 

 be the best of the media tested. 



Engberding gives 24 sets of plates; of these, 14 are of six plates each, 

 six of five plates, three of four plates and one of nine plates. Nearly all 

 the sets show excessive variability; only three values out of the 24 are 

 below the expected average for the corresponding number of plates. The 

 total of the 24 values is 5-36 times the expected total. No further test 

 is necessary; random sampling must be regarded as one of the smaller 

 causes of variation in these data. 



(C) The data of Breed and Stocking {13) 



We next come to a very thorough attempt made by Breed and 

 Stocking to test and improve the methods used in the bacterial analysis 

 of milk. The medium used in the platings here considered had the 

 following composition: — "Difco" peptone 1 per cent., lactose 1 per cent., 

 "Lemco" -3 per cent., air dried agar 1-5 per cent. A single batch of 

 medium was used throughout each experiment, so that abihty to re- 

 produce the medium, is not here tested. Parallel samples of normal milk, 

 and of milk inoculated with B. coli, were analysed by different analysts 

 and at different stations. Two series of these records have been examined 

 by comparing the different plates of each separate analysis. Each series 

 yielded 132 sets of three numbers, the duplicate counts of the same set of 

 plates being reckoned as two. If the duphcate counts had closely agreed, 

 this would tend to give us a bad fit between observation and expectation, 

 to the extent of doubling )^. Though the agreement is not sufficiently 

 great to have this effect, the tendency is to be borne in mind. 



The expected and observed distributions arc shown in Table XX. 



As with Buddin's data, though to a less extent, there is a small 

 systematic excess of the larger values; the mean variance in series B 

 is about 30 per cent, in excess of expectation, while in series C it is only 



