EVOLUTION 431 



span, and forearm of father, mother, and one to four 

 children. Now from these we can take out the statures of 

 1000 husbands and wives, and form a correlation table in 

 the manner indicated in the last chapter (see p. 400). If 

 we select from this table, given opposite, the husbands of a 

 given height, we have corresponding to them an array of 

 wives. Now if there be no assortative mating, no tendency 

 of like to seek like, then the mean of all these arrays of 

 wives ought to be sensibly the same, i.e. the mean height of 

 wives in general. But what do we actually find ? Why, 

 if the height of the husband is above the average, then 

 the average height of the array of wives sensibly exceeds' 

 the mean height of wives ; and if the height of the husband 

 is below the average, the average height of the array of 

 wives is sensibly below the average height of wives. In 

 other words, tall tends to marry tall and short to marry 

 short. There is a real correlation between the stature of 

 husbands and wives, and we can determine its value by 

 the regression line precisely as we did for the stigmatic 

 bands on the poppy capsules, p. 395. Doing this we 

 find •} — o ■ ^ 



or husband and wife resemble each other in stature far 

 more closely than we shall find uncle and niece do (p. 48 1). 



For eye-colour I was able to extract from Mr. Francis 

 Galton's family record data the colours for husband and 

 wife in 774 cases. The classification here takes place by 

 assigning each individual to one of the following eight 

 groups to which the numbers i up to 8 are attached : 

 (i) light blue; (2) blue, dark blue ; (3) blue-green, gray ; 

 (4) dark gray, hazel ; (5) light brown ; (6) brown ; (7) 

 dark brown ; (8) very dark brown, black. 



On the basis of this classification we have the follow- 

 ing table : — - 



1 For a totally different series embracing only 200 couples, measured a 

 number of years ago by Mr. Francis Galton, I find r=.093l. This value is 

 much lower, although still higher than the correlation of first cousins. It is 

 subject to a much larger probable error. Weighting with the offspring I found 

 ^=.1783 for 965 fathers and mothers, again a higher correlation than for 

 uncle and niece. 



