502 
GENETICS: W. E. CASTLE 
agencies are eliminated, such as differences of temperature, age, or 
associated genes. But by so doing, they introduce other and much 
more serious disturbing agencies. 
Take the case of the three genes, yellow, white and bifid, which 
both they and I have singled out as well illustrating the merits of 
our respective methods. Morgan and Bridges^ in describing their 
method of constructing 'Diagram I' say. ''Thus if the experiments 
give a cross-over value of 5 per cent for white and bifid, we say that 
white and bifid lie 5 units apart in the X chromosome. Other experi- 
ments show that yellow and white are about 1 unit apart, and that 
yellow and bifid are about 6 units apart. We can therefore construct 
a diagram with yellow as the zero, with white at 1, and with bifid 
at 6." But the Table 65 of this same publication, on the basis of 
which, it is explicitely stated. Diagram I was constructed, gives these 
distances more exactly as yellow-white, 1.1; white-bifid, 5.3; yellow- 
bifid, 5.5. The authors in the passage quoted ignore a considerable 
TABLE 1 
FULL 
DATA 
SELECTED DATA 
Cross-over 
values 
Cases 
Cross-over 
values 
Cases 
Yellow- white 
1.1 
81,299 
1.2 
1,218 
White-bifid 
5.3 
23,595 
3.5 
1,218 
Yellow-bifid 
5.5 
3,681 
4.7 
1,218 
discrepancy with these figures, since it is obvious that if the three 
genes are, as supposed, in a straight line, the distance yellow-bifid 
should equal the sum of yellow-white plus white-bifid, or 6.4 (instead 
of 5.5). Commenting on the actual observations in the case, I said, 
"Therefore bifid can lie neither above nor below yellow and white, 
in the line which joins them, but must lie laterally about equidistant 
from both.'' Returning to the case, Sturtevant, Bridges and Morgan 
now reject all but a small part of their previous data, retaining only 
two experiments in which all three loci were simultaneously under 
observation. The changed character of the data is shown in Table 1. 
Yellow- white is not materially affected but white-bifid is now 3.5, 
instead of 5.3, and yellow-bifid is 4.7, instead of 5.5. Are the new 
values more reliable than the old ones? So far as amount of data 
is concerned, they are much more liable to error through random 
sampling. If disturbing agencies are at work in isolated experiments, 
their effect is less likely to be felt in larger totals based on experi- 
