412 
Cooperative Investigations on Plants 
not group themselves between "4 and "5, the vahies found for characters in other 
types of life. We do not believe that experiments on plant heredity, which employ 
no measurement and allowance for these modifying factors, can lead to conclusive 
results. In our opinion these factors are all potent, and modify widely the limits 
of any category which may be selected for classification of simple characters in 
plants. 
(7) On Collateral Heredity. 
The following table gives the results obtained for the 1903 crops with a com- 
parison of such characters as are comparable in the Oxford crop of 1900. In all 
cases the apical capsules in the sibling plants are compared. In the Appendix 
are given the actual tables from which the results are deduced. It is desirable 
to point out the methods adopted in the individual cases : 
(a) Resemhlance in number of Stigmatic Bands. Found by usual correlation 
table. 
{h) Number of Petals. Fourfold table, the grouping being into the normal 
four petals and the abnormal more than four petals, including in the latter class 
semi-double, double flowers, and flowers with one or more petaloid stamens. 
(c) Colour of Middle Third. Fourfold table, the grouping being into flowers 
showing some and flowers showing no red. 
(d) Breadth of Margin. Fourfold table, the grouping being into flowers 
showing some and flowers showing no margin. 
(e) Colour of Flower at Margin of Petal. Fourfold table, the grouping being 
into flowers showing some and flowers sliowing no white there. 
(/) Extent of Basal Patch. Fourfold table, the divisions being into none or 
slight, and large or definite patches. 
(g) Wrinkling. Fourfold table, the divisions being into marked wrinkling 
(' frilled ' or ' wrinkled ') and insignificant wrinkling (slight or no wrinkling)*. 
Several points are brought out by this table. In the first place we see 
that the starveling Streatham crop shows a great reduction in the fraternal 
correlations. Now if we endeavour to interpret this result we should reasonably 
suppose that much selection took place at Streatham ; the mean of the stigmatic 
bands and their variability has been much lowered. There is further no reason 
for supposing a correlated selection in brother plants. Now if we suppose two 
organs A and B with standard deviations o-j and and correlation r^.2, to be selected 
in a non-correlated manner so that their standard deviations become s^ and s^, it is 
easy to show, on the hypothesis of nearly normal distribution, that the correlation 
after selection will be 
p — Si ^2 '^'12 ^-^ 
and this independently of changes in the means. 
* Shape of the capsules, for which only data at Crewe were available, gave an uncorrected corre- 
lation between sister plants of '15. See Table W. Appendix. 
