i64 THE FIRST PRINCIPLES OF HEREDITY 



Why should there be regression ? Does it agree with our 

 general notion of inheritance ? Though the law of regres- 

 sion seems to be something unexpected, yet it is completely 

 in accordance with the known facts of heredity. We have 

 found that in any given population the medium variations — 

 be they of height or any other characteristic — are the most 

 frequent. Now, the forefathers of any given son will, 

 according to the law of probability, be on the average 

 mediocre people, and as the son inherits, not only from his 

 parents, but from all his ancestors, it is, to use an expression 

 of Karl Pearson's, " the w^eight of this mediocre ancestry " 

 which drags down any exceptional gifts the son may 

 inherit from his parents. " But the law of regression," 

 says Galton, " is even-handed ; it levies an equal succession 

 tax on the transmission of badness as of. goodness. If it 

 discourages the extravagant hopes of a gifted parent that 

 his children will inherit all his powers, it no less dis- 

 countenances extravagant fears that they will inherit all 

 his weakness and disease." 



Correlation has been found to apply not only to stature 

 of man, but also to eye-colour, and even, as Karl Pearson 

 has shown, to mental and moral characteristics. The co- 

 efficient of correlation gives us therefore a convenient 

 measure of nearness of relationship between family 

 members. Thus, the Regression Table of Galton on p. 162 

 gives us at the same time a numerical value for the degree of 

 kinship between the relations mentioned. The coefficient 

 of correlation has therefore also been called the coefficient 

 of heredity. We see, e.g., that brothers are twice as closely 

 related to each other as father and son. 



/ V.—F UR THER CORRELA TIONS. 

 (a) ASSORTATIVE MATING. 



We have seen that there is correlation between parents 

 and children and all other degrees of kinsmen whose 

 relationship can be expressed in definite numerical values. 



