1918] Waynick: A Statistical Study of Nitrification in Soil 265 



While we are not justified in using the results obtained from labora- 

 tory treatments as direct criteria of what will happen in the field, it 

 is probable that greater variability will be found in field plots which 

 have been subject to treatment with fertilizers than in the normal soil 

 to which no fertilizers have been added, especially when the fertilizer 

 is of such a complex nature as dried blood. 



It must further be emphasized that the figures as given above apply 

 only to the soil under discussion ; each soil with the treatment applied 

 must be considered as a unit in a statistical study. It is evident that 

 a standard established for the residual nitrate, for instance, will not 

 hold for the nitrate produced from a one per cent application of blood. 

 The exact correlation between the various laboratory treatments and 

 between those treatments and results secured in the field is reserved 

 for discussion in a future publication. 



Representative and Composite Samples 



It is evident from the results already presented in various group- 

 ings that a representative sample of any soil is a purely hypothetical 

 quantity whose constituents may only be determined indirectly. It is 

 also evident that determinations reported from only one sample of a 

 given area are practically worthless, when the errors to which it is 

 subject are of the magnitudes shown above. Composite samples made 

 up from a small number of single samples are of little more value. 

 Not until enough samples have been taken to enable the proper calcu- 

 lation of the probable error of the mean of a given number of samples 

 is the making of a composite sample justified, since the results obtained 

 from working with such a sample are of very limited value unless the 

 error to which the composite sample, itself, is subject, can be deter- 

 mined. 



It seems, therefore that the use of the word composite as relating 

 to soil samples is only justified when the variations to which the 

 individual samples are subject are known and enough samples taken 

 to establish the error to which the composite, as a mean of all the 

 samples, is liable. The magnitude of the error which is liable to creep 

 into the determinations due to imperfect mixing in a composite sample 

 is no doubt worthy of consideration, but since its value is undetermined, 

 it has been assumed that a composite actually represents the mean of 

 all the samples taken. Just how reasonable such an assumption may he 

 remains for future investigations to bring out. 



