A. Baiirington, a. Lee and K. Pearson 
259 
of brothers or about ll'o pairs to each pair of parents; brothers of the same 
litter from the stud-book were given by 532 pairs of parents and provided 2144 
fraternal pairs, or about four pairs to each pair of parents ; lastly, brothers of 
different litters from the stud-book were provided by 258 pairs of parents, and 
these gave 1044 pairs of brethren, or again about four to each pair of parents. 
Could it be that this difference in the number of brethren to each pair of parents 
was the source of the large drop in correlation as wc passed from schedules to 
stud-book? This point was then taken up and investigated*. The members of 
a litter were given numbers, 1, 2, 3, ... and these were put on tickets and pairs of 
these tickets drawn at random. In this way 2086 pairs of brothers were obtained, 
no brother being used with more than one other brother. The table thus obtained 
is given as Table MMMM in the Appendix. The correlation obtained by the 36-fold 
table is '6697, while for 7484 bi'others using every brother with every other 
brother it is '6607. The difference is in fact insensible, but the larger correlation 
is in the series of non-repeated brothers. We conclude therefore that the difference 
between the fraternal correlation for brethren from the same litter as taken from 
the schedules and from the stud-book is not due to any purely arithmetical effect 
of taking many pairs from large litters. The only other explanation we can give, 
but we see no means at present of testing its truth, lies in the view that the 
colours of the litter were taken at a much earlier stage for Mr Howard Collins' 
schedules, than for the stud-book. In the latter cases dogs are often not entered 
for a year or two. It is possible therefore that the litter as a whole is more alike 
TABLE VIII. 
Collateral Heredity. Resenihlance of Siblings. 
Unselected Kecokd 
Selected Eecoud 
Nature of 
Sibling Pail- 
Same Litter, 
Repeated Brothers 
Same Litter, Non- 
repeated Brothers 
Same Litter 
Different Litters 
No. 
Correlation 
No. 
Correlation 
No. 
Correlation 
No. 
Correlation 
Dog and Dog . . . 
Bitch cand Bitch 
Dog and Bitch 
7484 
6542 
7175 
•661 
•700 
•669 
2086 
•670 
2144 
2002 
2864 
•524 
•595 
•558 
1044 
1010 
1031 
•521 
•558 
•509 
Mean 
•676 
•670 
•559 
•529 
* The whole problem is a difficult one, there are so many statistical pitfalls surrounding it. 
Theoretically it would seem that both for fraternal and homotypic correlations the increase in the 
number of brothers or of homotypes taken from each unit ought to alter the correlation, actually we 
cannot find much difference when the same material is worked out with many and with few members of 
each unit. See the " Homotyposis" memoir, Phil. Trans. Vol. 197, A, pp. 310 — 313 ; the memoir on 
"Mendel's Principles," Pldl. Trans. Vol. 203, A, pp. 74—77 and the paper on " Telegony in Man," 
R. S. Froc. Vol. GO, p. 279. 
83—2 
