J. A. Harris 
471 
Studying the tables first for the prevalence of positive and negative signs 
we find the results : 
B— A, 11 negative and 17 positive differences, 
C — B, 18 negative and 10 positive differences, 
C — A, 19 negative and 9 positive differences. 
Taking up next the significance of the differences for the constants for indi- 
viduals as measured by their ratio to their probable error, we find results as follows. 
For the comparison B — A only 12 of the 28 differences are significant as com- 
pared with 18 out of 28 in the case of the means. Of these 6 are positive and 
6 negative in sign. The difference C — B shows 20 significant constants, of which 5 
are positive and 15 negative. C — A shows 17 significant constants, of which 4 are 
positive and 13 negative. The first comparison indicates that selective elimination 
has no influence upon variability. The second and third both suggest that the 
variability is sensibly less after elimination has taken place. 
The differences in the variabilities for the lumped collections may now be 
examined in Table IX. 
TABLE IX. 
Comparison 
Difference 
and 
Probable Error 
Diff. 
E dm. 
B— A 
+ -0756 + -0091 
8-31 
C— B 
- -09.38 + -0086 
10-91 
C— A 
- -0182+ -0081 
2-25 
At first these results puzzled me very greatly; and the following interpretation 
is offered only tentatively and as a working hypothesis for further investigation. 
The collection B probably comes the nearest to representing the original popu- 
lation of ovaries of any that we have, since it still contains a high percentage of 
ovaries which will be eliminated later*. 
Now it is quite clear that if the ovaries with the lower numbers of ovules per 
locule are eliminated, their variability will be expected to be less than that of the 
original population. Likewise if the ovaries with the higher numbers of ovaries 
develop to maturity it is clear that their variability also should be less than that of 
the population as a whole. In short, both the samples of eliminated ovaries and 
the samples of matured fruits represent the more extreme portions of the curve of 
variation. If this be true, we should expect to find the variability of B greater 
than that of A or C, while an inevitable consequence of this condition of affairs 
* I am sorry not to be able to give definite quantitative data bearing upon this point, but this 
is quite out of the question in a ligneous plant like Staphylea, where the number of tlowers produced is 
immensely large and where it is impossible to determine with certainty from the mature inflorescence 
the number of flowers originally produced. 
60—2 
