256 On Inlieritance of CiKit-Colour in the Greyhoimd 
of parent and offspring is sensibly reduced by selection in recording. Mr Howard 
Collins' data enable us, however, to roughly allow for this selection ; the corrected 
values are indicated in the last column of Table IV, and we shall proceed to use a 
like correction in the following paragraph. 
(G) On Direct Inheritance. Grandparental Correlations. We can here 
unfortunately only deal with the data from the stud-books. The greyhounds, 
whose offsprings were recorded for Mr Howard Collins, appear to be in great bulk, 
if not entirely, well-known hounds whose names are in the stud-books. The 
selection arises from only one or two individuals in each litter being recorded in 
the stud-books. This, as we have seen, weakens the correlation ; possibly it arises 
because the offspring of more fashionable colours only are recorded in the case 
of a litter, the pai'ents of which may be of less fashionable colours. If we consider 
grandchildren of stud-book parents, then there has clearly been a selection out 
of the litters of this generation, not only of some members of each litter, but of 
whole litters, because certain parents have been omitted from record altogether, 
as the rejected members of the litters of the first generation. Now the problem 
thus stated becomes undoubtedly a complex one, as the reduction for influence 
of selection on correlation, if the selection extends over several generations, is by 
no means easy to determine. But it seems not unreasonable to suppose that the 
total distribution of colour in the sporting greyhound, although probably subject 
to secular change, is not changing very rapidly. In other words, the effect of 
selection for breeding is to maintain a nearly constant colour distribution for a 
few generations. If this be so, the chief effect of selection in record is due to 
the weakening of the correlation by selection of off'spring and not to the selection 
of the breeding parents or grandparents. Accordingly it seems to us that all 
we can do is to divide the grandparental correlations we find from stud-book 
pedigrees by the factor -875, and treat these as the best approximations available 
to the grandparental correlations. We have then the following results. 
TABLE VI. Grandparental Correlations. 
Mean Square Contingency 36-fold tables. 
Grandparent and Offspring 
Number 
Kaw 
Value* 
Probable 
Eeduced Value 
Paternal Gi-andsire and Dog 
952 
•314 
Paternal Grandsire and Bitch ... 
893 
■229 
•261 
Maternal Grandsire and Dog 
939 
■264 
■301 
Maternal Grandsire and Bitch . . . 
888 
•336 
•384 
Paternal Grandam and Dog 
947 
•322 
■368 
Paternal Grandam and Bitch 
900 
•272 
•311 
Maternal Grandam and Dog 
911 
•291 
•332 
Maternal Grandam and Bitch ... 
863 
•338 
•386 
Mean 
•291 
•,332 
* The grandparental tables treated by the fourfold method gave— owing to some extent to paucity of 
number —rather irregular results. The mean of 16 fourfold tables Bed A and Black B was •213, and of 
sixteen IG-fold and 25-fold contingency E and F tables •247, both less than the 36-fold tables. 
