FILIAL REGRESSION 317 



Grammar of Science (igoo, p. 454). " Fathers of a given height 

 have not sons all of a given height, but an array of sons of a 

 mean height different from that of the father and nearer to the 

 mean height of sons in general. Thus take fathers of stature 

 72 inches, the mean height of their sons is ^0"-^, or we have a 

 regression towards the mean of the general population. On 

 the other hand, fathers with a mean height of 66 inches give 

 a group of sons of mean height 68""3, or they have progressed 

 towards the mean of the general population of sons. The 

 father with a great excess of the character contributes sons with 

 an excess, but a less excess of it ; the father with a great defect 

 of the character contributes sons with a defect, but less defect 

 of it. The general result is a sensible stability of type and 

 variation from generation to generation." 



The quotations which we have given make the general idea 

 of regression quite clear ; for the detailed evidence and for 

 further elaboration we must refer to the works of Galton and 

 Pearson. 



It is necessary, however, to ask what this statistically 

 established fact of filial regression really means biologically. 



Interpretation of Regression. — The facts of regression 

 are expressed as a whole in the striking statistical resemblance 

 between successive generations of a people. There is a continual 

 tendency to sustain the specific average. It can hardly be 

 denied that the similarity is in part the result of similar con- 

 ditions, e.g., of selection, but this hardly applies to the proportions 

 persisting between tall and short, dark and fair, and so on. 

 That it is not . due to completeness of inheritance is obvious, 

 for " the large do not always beget the large, nor the small 

 the small " ; the children do not in any precise way repeat the 

 qualities of their parents. (Galton, 1889, pp. i and 116.) On 

 what then does this regression depend ? 



Galton suggests two different reasons for the occurrence of 

 regression (pp. 104, 105). The first is connected with his idea 



