EXPERIMENT STATION BULLETINS. 



459 



Series VI. 



Average Errors. — Since these data will be used for a very extensive 

 discussion, it is necessary to realize how large the probable error is in 

 order to avoid conclusions which have no other basis than an analytical 

 error. Of the 540 duplicate determinations, only seven showed such a 

 difference that they were excluded from the averages, or rather that only 

 the probable datum was taken while the other one was discarded. There 

 are three in series iii. ; three in series iv. ; and one in series vi. They are 

 considered by the author as due to contaminations or incomplete sterili- 

 zation, though no proof can be given, since the deviation could not be 

 discovered until the distillation was finished. The grand average error 

 of all other 533 duplicates is ±0.25 cc. of n/lO NH3. The averages are 

 simple arithmetical averages; the method of the least squares was not 

 used because of the amount of work involved. 



The error is largely due to the analytical method, and only to a slight 

 degree dependent upon variation in the microbial growth, since it in- 

 creases but very little with the increasing amounts of ammonia. 



Average Error of All Determinations. 



