H. H. Robinson — Chemical Analyses of Igneous Rocks. 271 



summations are more uniformly distributed between his two 

 maxima. However like D, he has more summations in the 

 lower maximum group than in the upper one. Analysts F and 

 G have distributions which are quite similar, although F's 

 summations lie largely on the negative side of the general axis 

 (100*13 per cent), whereas those of G are largely on the posi- 

 tive side. 



The foregoing double probability curves may be resolved 

 into their two component single probability curves and it then 



Fig. 5. 



ffJZ 



<f<?.37 ff6Z qfS-7 



/OOJZ 



10037 



/OO. 62. 



/Q0&7 /O/ 12. 



appears that the two parts may be classified under the three 

 heads noted in an earlier paragraph, practically all falling 

 under (2) and (3). 



A distribution of summations to be significant must rest, of 

 course, on sufficient data, and this is essentially true for the 

 examples already cited. The number of analyses necessary to 

 give such a result will vary widely, depending upon the 

 analytic skill of the individual chemist. The minimum num- 

 ber can be determined approximately by examining the work 

 of different analysts. In fig. 5 are given four examples to 

 show that a very good distribution of summations, as indicated 



