232 Castle. 



should be observable on one of the hypotheses stated by Bateson in 

 the passage last quoted. 



No one will, I think, question the view that size differences involve 

 many distinct physiological processes within the organism, which are 

 to a greater or less extent independent of each other, or at least in- 

 dependently variable, but this is very different from assuming that each 

 of these processes constitutes a Mendelizing unit or factor, yet the 

 current explanation of size inheritance assumes that only such units 

 or factors are concerned in size inheritance, a wholly unwarranted sup- 

 position, it seems to me. For suppose that instead of regarding Mendelian 

 gametes as pure, we regard them as capable of contamination in crosses, 

 an hj'pothesis supported by a large amount of experimental evidence. 

 Suppose furtlier that we cross two races of animals one of which has 

 twice the average size of the other, as Phillips did when he crossed 

 Rouen and Mallard ducks. We may for simplicity call the mean size 

 of one race 3 and that of the other 6. If Fi is strictly intermediate 

 it will have a size value of 4'5. The experimental evidence shows 

 unmistakably that in such cases the gametes formed by the Fi individuals 

 transmit this same size value, though with slight plus and minus 

 variations, so that the Fa generation formed by combinations of such 

 gametes among themselves is somewhat more variable than Fi. It is 

 conceivable that these plus and minus variants arise by contamination, 

 the 3 and 6 gametes which united to form Fi mutually influencing each 

 other so that the original 3 gamete emerges from the cross with an 

 increased value of 4, and the original 6 gamete emerges with a decreased 

 value of 5. Gametes of such modified values (4 and 5) uniting inter se 

 would produce the following array: 



Values 4 4-5 5 



Frequencies .... 1 2 1 



Such an Fa distribution would show greater variation than that 

 of Fl (assumed to be all 4'5). The variation would be about the same 

 mean and might well be characterized as a blend with increased vari- 

 ability. Selection of extreme variates from the Fs array would also be 

 effective, since those of value 4 might be expected to produce an Fg 

 population of lower mean value and on the whole less vai'iable than the 

 population produced by an F» of value 4'5, itself jiroduced by the 

 dissimilar gametes 4 and 5. 



Now I do not mean to assert that modifications induced by crossing 

 are as definite and regular in occurrence as those assumed in the 

 imaginary case just described. Such an assumption would accord better 



