414 
Cooperative Investigations on Plants 
most probably be accounted for on the basis of selection, such selection really 
reducing the correlations in a manner qualitatively and, as far as can be roughly 
determined, quantitatively in the manner observed. 
In the next place, we have here made no allowance for possible selection of 
maternal plants and none for homotyposis. Hence the value of our fraternal cor- 
relation is again a minimum value. Even including the Streatham results the 
average correlation for all characters is -23 and without these it is -26. The chief 
irregularities are in the values of the correlations for wrinkling, extent of basal 
patch and margin, where we should expect minute differences of soil or situation, 
time of observation or personal equation to be more influential*. Now we might at 
first argue that '25 would be a satisfactory value to reach for fraternal heredity, as 
being about half the value noted for fraternal correlation of whole brothers, but 
there ai'e three points to be considered here, (i.) The neglect of any correction 
for selection, whether of maternal plants or of surviving offspring, has certainly 
given too low a value to these correlations, (ii.) There is a further correction 
to be made for the homotyposis which remains after allowing for differentiation of 
the first flower. This is probably not substantial in some of the characters, but it is 
considerable in others. It would again tend to raise the observed values, (iii.) In 
considering the result we have assumed that daughter plants grown from the 
same capsules are sisters, or that the pollen parents are different for each pair 
of cases. We have already indicated (p. 398) tliat we consider that there is a fairly 
wide cross-fertilisation, i.e. a considerable variety of pollen parents to plants grown 
from the same capsule, but it is hardly likely to be complete for each pair of sister 
plants, and will probably vary considerably from capsule to capsule. Thus half- 
sibling correlation is rather a minimum limit to what we might anticipate, and not 
what is actually likely to occur. 
Hence the apparent reasonableness of a value about "25 is seen to be rather 
fortuitous. 
Lastly, granted that the sister plants are in the bulk half-sisters, and that the 
average value of the correlation of whole sisters is about "5, it appears open to con- 
siderable questioning whether we ought really to expect that the half-sibling 
resemblance will be just half the whole-sibling resemblance. This may appear to 
some to flow at once from a, ipriori considerations, but we do not feel so clear on 
this point now as a few years ago ; for the determinations of whole- and half-sibling 
relationships which have been made so far do not bear it outf. 
* Experiments at University College on personal equation show a continuous change during a long 
series of observations. Judgment would thus depend on the time of flowering, and brother plants not 
being registered usually on the same occasion, it follows that personal equation would be likely to 
introduce a small spurious correlation between non-brothers, rather than to emphasise the degree 
of resemblance between brothers. 
t As in the case of all vital problems, however, the complexity is very great. Horse and cattle 
breeders usually inbreed to some extent, and it very often happens that two half-siblings have more 
than four common great-grandparents, and so should be more alike than normal half-siblings with 
only four in common. 
