EVOLUTION 427 



Difference in type between husbands and males = - .079 + .142 

 „ variability „ ,, =+.036 + .101 



,, type between wives and females = —.174 + .126 



„ variability „ ,, =-.025 + .085 



The numbers here given with db signs are the " prob- 

 able errors," or the quantities by which the mathematician 

 tests the probable divergencies of the true from the observed 

 value. In the four differences of type and variability, there 

 is only one in which the difference exceeds the probable 

 error of the difference. We cannot even in this case assert 

 a significant difference between the stature of wives and 

 women in general ; indeed, if there were one it might 

 possibly be accounted for by causes other than preferential 

 mating. Our statistics, however, run to only a few 

 hundreds, and were not collected ad hoc. Still, so far as 

 they go, they show no evidence of preferential mating in 

 mankind on the basis of stature, or of any character very 

 closely correlated with stature. Men do not appear, for 

 example, to select tall women for their wives, nor do 

 they refuse to mate with very tall or very short women, 

 for then wives would be less variable than women in 

 general. 



In the next place let us consider eye-colour ; here I 

 have much more numerous data, and want of homogeneity 

 in age makes itself less felt. The following results are 

 based upon very substantial numbers, but as there is no 

 quantitative scale for eye-colour peculiar methods had to 

 be adopted for their treatment. The material was taken 

 from Mr. Francis Galton's records, kindly placed at my 

 disposal. He classifies eye-colour into eight groups. One 

 of these is blue-green to gray, and within this group the 

 type always falls. Accordingly the range of blue-green 

 to gray was considered as unit length on the eye-colour 

 scale, and types are stated as the fraction of this range 

 they are from its lighter end. Further, all variabilities 

 are measured by the multiple the standard deviation is 

 of the same range. This being indicated we have the 

 following system : — 



