426 
R. A. FISHER ON THE CORRELATION BETWEEN 
and making use of the fraternal correlations to separate c x and c 2 , by the equations 
/ i c if.l + c 2 (l + 2A)], 
c 1; =4/-2p-/x, 
1-0333 -8139 -078 
■6864 -7842 077 
•3883 -2850 -105 
The standard error for the dominance ratio is now very high, since the latter is 
proportional to the difference f—p. If we assume a known value for c 1; and calculate 
the dominance ratio from p and m only, the standard error falls nearly to its value in 
Article 18. 
The three values for the ratio of the ancestral correlations '691, '640, '655 are 
now higher than that obtained from observations of eye colour, and are more similar 
to the value '660 obtained for the coat colour of horses. Without knowing the 
marital correlations in these cases, it is not possible to press the comparison further. 
It would seem unlikely that the conscious choice of a mate is less influenced by eye 
colour than by growth features, even by stature. But it is not at all unlikely that 
eye colour is but slightly correlated with other features, while the growth features 
we know to be highly correlated, so that a relatively slight selection in a number of 
the latter might produce a closer correlation in each of them than a relatively intense 
selection of eye colour. 
The value of Cj for span is still greater than unity, 1'033, but no longer unreason- 
ably so, since the standard error is about '078. If we were considering span alone the 
evidence would be strongly in favour of our third hypothesis. A remarkable con- 
firmation of this is that Pearson and Lee ( loc . cit., p. 375), considering organic and 
marital correlations alone, show that the observed correlations could be accounted 
for by the following direct selection coefficients : — 
Stature. Span. Cubit. 
■2374 -0053 -1043 
Naturally these cannot be taken as final, since there are a large number of other 
features, which may be connected with these and at the same time may be subject to 
sexual selection. The correlations of cross assortative mating are in fact smaller 
than they would be if direct selection to this extent were actually taking place. The 
influence of other features prevents us from determining what proportion of the 
observed association is due to direct selection, but if inheritance in these growth 
features is capable of representation on a Mendelian scheme — and our results have 
gone far to show that this is likely — it would be possible to distinguish the two parts 
by comparing the parental and fraternal correlations with those for grandparents 
and other kindred. 
On our present supposition that the association is primarily in z, and for the case 
or 
we obtain 
c t . . -8796 
c 2 . . -8331 
~ . . -2450 
