6 14 C. B. Davenpcrt and Marian E. Hubbard. 



30 rays and an index of 2. The higher index indicates the higher 

 potentiality in evolution. 



On the other hand the coefficient of variability in integral vari- 

 ates marks rather the inherent capacity to vary — not of the system 

 as a whole — but of each of its numerically repeated parts. It 

 disentangles, as it w^ere, the variability due to complexity or mani- 

 foldness of structure and the inherent variability of parts. Com- 

 paring the two we may say the index is concerned only with the 

 end-result and makes no allowance for the complexity of the con- 

 tributory causes; it measures, in a way, the complexity combined 

 with the variability of the germ-plasm. The coefficient, on the 

 contrary, more nearly measures the relative variability of the germ- 

 plasm alone. 



The idea of the difference between complexity combined with 

 variability of parts and variability alone may be assisted by an illus- 

 tration. Imagine two projected railway lines of which the profiles 

 are to be obtained by leveling. The one is very short, the other 

 long. In leveling the shorter course, even though the level has to 

 be set up only once, there are certain errors. These will be largely 

 in the instrument itself due to its imperfect adjustment, or they 

 may be due to errors in the graduation of the rod. These may be 

 called the constant errors; they correspond to the inherent vari- 

 ability of the germ-plasm. In the longer course, where the level 

 has to be set up many times, a different sort of error has to be 

 taken into account. This is the sum of errors due to imperfect 

 reading of the rod on the turning points, due to irregular heating 

 of the sides of the telescope, due to diffraction of light by the 

 atmosphere. This second sort of error is mostly outside of the 

 instrument itself, and it tends to 'ncrease with the length of the 

 course, and to a certain extent with its irregularity. Now, if two 

 equally accurate leveling parties were getting the profile between 

 two pairs of points of which the elevations are known, the distance 

 of the one course being short and the other long, we should expect 

 the absolute error at the end of the long course to be greater than 

 at the end of the short course. In comparing the accuracy of the 

 two parties we should feel justified in dividing the absolute error 

 made by each by the length of the line traversed — expressing the 



