340 



Method of estimating Bacterial Density 



5-plate than in the 4-plate sets, possibly because the sets of five plates 

 make possible a closer scrutiny into the exactitude of the agreement 

 between the observed sets, and samples from a Poisson Series. 



For the same reason the 50 sets of three plates cannot be expected 

 to provide much additional information. The seven exceptionally high 

 values stand out perfectly clearly; the lowest is 9-2, a value which would 

 be exceeded by only one normal sample (of 3) in 100. The next highest 

 values 5-4 and 6-4, would not be suspect save for their occurrence in 

 December; they will be treated as normal. 



Since the 3-plate sets are relatively scanty, we can best test their 

 agreement with theory by dividing the theoretical distribution of 43 

 values at its quintiles, so that the expectation is the same in each group. 

 We then have 



Table X. Sets of three plates 

 , x^ = 1-77 P = -^To 



Total 



The agreement with expectation is excellent, and the sets of three 

 plates bear out the conclusions derived from the sets of four and five 

 plates, save that here there is no visible excess of Ioav values of x^- 



It appears therefore that out of the 362 sets of plates examined the 

 majority represent true samples from the Poisson Series, such as would 

 be the case if the biological and technical difficulties of the bacterial 

 count method as appHed to soil had been completely surmounted. Forty 

 sets, which can be identified almost with certainty, are affected by some 

 cause or causes which greatly increase the variability between the plates, 

 Avhile probably a smaller number, including apparently none of the 

 3-plate sets, are affected by a second cause of error, which reduces the 

 variabihty between the plates. 



