126 Experimental and Statistical Studies upon Lepidoptera 
Furthermore reversal of selection with respect to variability appears in some 
cases with astounding clearness. The perfect survivors are cer'tainly more variable 
in total length, probably so in bust length, possibly in weight, bust width and bust 
depth. No significance attaches to the differences in bust proportions. While 
they are possibly more variable in antennal length, and possibly less variable 
in antennal width, the perfect survivors are certainly more variable in antennal 
proportions. 
These results stand in the sharpest contrast to those referring to pupal 
elimination. As the proportionate characters of the bust are involved no less 
than the absolute values it must be concluded, I think, that the process of pupal- 
imaginal elimination involves certain factors which do not operate during pupal 
existence. Hence individuals which could and did survive the conditions of purely 
pupal existence find themselves marked for destruction by the very characteristics 
which formerly proved advantageous. 1 regard the reversal on the basis of 
variability as indicating the same conclusion also. Although discussion of these 
points may properly be deferred, it may be stated here that the enquiry into the 
actual basis of the selective processes may lead to a clearer understanding of the 
apparently conflicting facts. 
V. T!ie Comparative VariaJyility of Males and Females. 
The question as to which sex is the more variable is of considerable interest in 
some connections. It is not directly related, howevei', to the px'oblem which is 
treated in the present paper, so that the facts may be presented in a brief form 
without any discussion at this time. 
It is clear that it is not possible to determine the relative variability of the two 
sexes by a simple comparison of the indices of variation for any given character, 
for with a greater absolute value of any measurement a higher degree of variability 
is usually associated. A more satisfactory basis for comparison is afforded by the 
Coefficient of Variation, which is found by dividing the index of variability for any 
character by the mean value of that character. In order to obtain a convenient 
whole number, this quotient is generally multiplied by 100. By this means 
a measure of variability is derived which is independent of the absolute values 
concerned, though certain mathematical defects, which need not be considered 
here, may attach to this process. 
In the subjoined table (Table X.) are given the coefficients of variability of the 
ten characters of both sexes, for the whole group (eight characters), for the 
surviving pupae, and for those which metamorphose perfectly. It will be seen 
that when the whole group is considered, the males are on the whole more variable 
than the females, the latter being more variable only in two of the bust measures. 
Among surviving pupae the male sex is far more variable, being less so in antennal 
proportions solely. And finally the same sex is the more variable on the whole as 
regards the body in the third group of perfectly metamorphosing pupae; for 
