H. H. Robinson — Chemical Analyses of Igneous Rocks. 261 



a lower figure because rocks which contain large amounts of 

 iron-bearing minerals are commonly rich in this element. The 

 results are as follows : 



Oxide determined Number of anals. Ave. summation 



Ferric oxide _ 96 100-13 



Ferrous oxide 21 100 04 



The average summation for the analyses in which only ferric 

 oxide was determined is the same as the general average, 

 whereas it would be expected to be distinctly higher. An 

 additional plus error due to magnesia would also be expected 

 as a number of the analyses are of inferior quality. The fail- 

 ure of the average summation to show a positive error is no 

 doubt largely due to the actual presence of little, if any, fer- 

 rous iron. The basic rocks used in this calculation are espe- 

 cially susceptible to alteration, actually shown by the large 

 amount of water many of them contain, and the ferrous oxide 

 has been raised to ferric form. The average summation for 

 the analyses in which only ferrous oxide was determined is 

 below the general average, an error in the right direction but 

 hardly of significant size unless the summation has been raised 

 by the plus error of magnesia. The result of this test is con- 

 siderably vitiated by the inferior quality of some of the analyses, 

 taken in connection with their small number. For example, 

 «two analyses of the same rock, by the same analyst, have sum- 

 mations of 99*15 and 100*51 per cent. This brings out the 

 point, which will be considered later, that the personal errors 

 of the analyst are sufficiently great to mask any error arising 

 from good analytic methods when the number of analyses 

 averaged is small. 



It is possible that the minus error in silica finds expression 

 in the silicic group of rocks (granites and rhyolites), which on 

 the average contain about 70 per cent of this oxide. For 

 example, 135 superior analyses of rocks in this group, selected 

 at random, have an average summation of 100*04 per cent as 

 compared with 100*13 per cent for the general average. The 

 difference may be explained by assuming an actual loss of 

 silica. Such a loss is not unlikely, because many of the 

 analyses, on which the above average is based, were made 

 before the present refinements in the determination of this 

 oxide were introduced. 



The foregoing considerations indicate that by far the 

 commonest errors which affect the summation of analyses are 

 of the analyst's own making. The trouble is not so much 

 with methods as with the analyst's ability to use them. Such 

 errors are illustrated by the results of a single chemist's work 

 on rocks of nearly identical composition. The following 



