466 
Selective Elhnination in StapJiylea 
respect to their probable error. In 18, 25 and 26 cases out of the 28 the differ- 
ences exceed 2'5 times their probable errors, and often reach many times their 
probable error. I think relationships of significantly positive to significantly 
negative differences, such as 13 : 5, 22 : 3, 25 : 1 can leave no doubt in the mind of 
the most sceptical that there is a very potent selective elimination, by which the 
mean number of ovules per ovary is very materially increased. 
The combination of the material from all the individuals gives the results for 
the differences in the mean number of ovules per locule presented in Table V. 
TABLE V. 
Comparison of Means for total Material. 
Comparison 
Difference 
and 
Probable Error 
Katio of Difference 
to its 
Probable Error 
Percentage Increase 
in Mean 
B-A 
■1426+ -0129 
11-054 
1-971 
C -B 
•3693+ -0120 
30^775 
5-005 
C-A 
■5119+ -0121 
42-306 
7-075 
Here the absolute differences are given in the first column, and the percentage 
increase of the higher over the lower mean in the last column. The results 
emphasize the conclusions drawn from the constants for the individual trees. 
The difference between the eliminated sample and the sample of ovaries 
remaining on the tree when they were taken is only about 2 per cent, of the 
former, but the difference is about 11 times its probable error and thus certainly 
significant. 
Doubtless the very slight divergence between the series A and B is due to the 
presence of a large number of ovaries in B that would a few days later fall from 
the tree and be classed with the A's, for we note a much greater increase from B 
to C than from A to B, amounting to no less than 5 per cent. 
As is to be expected the difference between the eliminated and the matured 
fruits is greatest of all, the latter having about 7 per cent, more ovules per locule 
than the former. 
The reader will note that in the last two comparisons the differences are 31 and 
42 times their probable errors. 
As a further demonstration of the differences in number of ovules per locule or 
per fruit in the three series I give (Diagram 2), the data in the form of an integral 
polygon in which the frequency on each grade is the sum of its own frequency and 
that of all lower grades. This form is one of the most convenient for purposes of 
comparison. 
It is unnecessary to table the actual frequencies for they can easily be com- 
puted by anyone who requires them from the totals of our tables. The percentage 
