424 
R. A. FISHER ON THE CORRELATION BETWEEN 
not justified for this feature. Accepting the above results for stature, we may ascribe 
the following percentages of the total variance to their respective causes 
Ancestry . 
Variance of sibship : 
£r 2 . 
Other causes 
Again it may be divided : 
Genotypes (a- 2 ) : 
Essential genotypes (r 2 ) 
Dominance deviations (e 2 ) . 
Association of factors by homogamy 
Other causes ... 
54 per cent. 
31 per cent. 
15 „ 
100 per cent. 
62 per cent. 
21 „ 
83 per cent. 
• 17 „ 
100 per cent. 
These determinations are subject, as we have seen, to considerable errors of 
random sampling, but our figures are sufficient to show that, on this hypothesis, it 
is very unlikely that so much as 5 per cent, of the total variance is due to causes 
not heritable, especially as every irregularity of inheritance would, in the above 
analysis, appear as such a cause. 
It is important to see that the large effect ascribed to dominance can really be 
e 2 
produced by ordinary Mendelian factors. The dominance ratio, -g, which may be 
determined from the correlations, has its numerator and denominator composed of 
elements, and a 2 , belonging to the individual factors. We may thereby ascertain 
certain limitations to which our factors must be subject if they are successfully to 
interpret the existing results. The values of the dominance ratio in these three cases 
are found to be : 
Stature. 
Span. 
Cubit. 
Standard Error. 
Dominance ratio . 
•253 
■274 
•336 
•045 
22. The correlations for uncles and cousins, still assuming that the association of 
factors is due to a direct selection of the feature x, may be obtained by the methods 
of Article 14, using the two series already obtained : that for ancestors 
c 1 c 2 
and that for collaterals, like sibs and double cousins, which have all their ancestors of 
a certain degree in common, 
ie 1 [l+c 2 (l + 2A)], 
1 V 1 [1 + 3c 2 (1+4A)], 
and so on. 
