J. A. Harris 
487 
For 4-merons fruits : 
Series A = 1-376 per cent. 
Series B = 2-248 per cent. 
Series C = 2-461 per cent. 
B— A = + -867 per cent. 
C— B = + -218 per cent. 
C—A = + 1-085 per cent. 
No probable errors have been calculated for the reason that our percentages 
are so low that the familiar theory of probable errors cannot be applied. 
It is quite impossible to say, therefore, whether or no these differences are 
really significant, but so far as our data go they indicate that the 2-merous ovaries 
are more likely to be eliminated than the 3-merous, while the 4-merous ones are 
less likely to be eliminated. 
V. D1SCUS.S10N OF Problems on the Basis of Actual Data for 1906. 
Problem 1. The seriations of the number of ovules per locule are found in 
Table XXXVIII ; those for total ovules per fruit in Table XXXIX. 
For constants we get : 
Series A ( 6—10 mm.). Mean = 7^232 + -029. 
Series B (11—15 mm.). Mean = 7-660 + -014. 
Series C (16—20 mm.). Mean = 7-821 ± -016. 
B— A = + -428 ± -032. 
C— B = +-161 ±-021. 
C— A = + -589 ± -033. 
The increase from A to B is 13-4 times its probable error, that from B to C 
is 8-6 times its probable error and the difference between the mean of the smallest 
and the largest fruits is 17-8 times its probable error. The increase from the 
smallest to the largest fruits is 8-14 per cent, of the value for the former. 
These results certainly*substantiate in the most unequivocal way those secured 
for means from the 1908 series, and are the more interesting from the fact that 
the material was collected not merely for a different purpose and with no idea of 
testing selective elimination, but by a quite different method. 
Problem 2. The absolute and relative variabilities for both ovules per locule 
and total ovules per ovary may again be considered. The complete answer to this 
problem is given in Table XXII. 
As the fruits become larger the variability becomes less. For the absolute 
variabilities the probable errors show that there is a very high probability that 
every difference is significant. Probable errors have not been calculated for the 
coefficients of variation, but the differences are all relatively large and consistent. 
62—2 
