206 BOTANICAL GAZETTE [MARCH 
and the number of seeds developing is greater than the correlation 
between the length of the fruit and the number of ovules formed.” 
III. Analysis of data 
We first examine our data to ascertain whether there is a measur- 
able degree of interdependence between the length of fruit and the 
number of ovules formed, and between the length of fruit and 
the number of seeds developing per locule. 
A. DISCUSSION OF DATA FROM INDIVIDUAL SHRUBS 
Perhaps the criticism, in these days, most generally directed at 
biometric work is that in the massing of large numbers of individuals 
into correlation tables, biological relationships which otherwise 
might be recognized are obscured. “The fundamental requisite 
for the validity of biometric constants is the homogeneity of the 
material upon which they are based,” we are told. There is more 
than one word to be said on this point, but certainly it is always 
desirable to consider data in as minutely analyzed form as can be 
done without incurring too great dangers incident to the probable 
errors of random sampling. 
From PEarson’s investigations’ we know that plant individuals 
of the same race are somewhat differentiated among themselves 
with respect to the characters of the organs which they produce. 
In short, they are individual really as well as nominally. The com- 
bination of series of fruits taken from a small number of plants 
might influence to some extent the correlation constants describing 
the relationship between the fertility of the fruit and its length. 
To free our results as far as possible from any such source of error, 
I first consider the relationship of the number of ovules formed (0) 
and the number of seeds developing (s) to the length of the fruit (I) 
in each of the 20 shrubs of the 1906 series, and each of the 16 of 
the 1907 collection. In the large series it is also of interest to com- 
pare the correlation for the length of fruit and number of ovules 
failing to develop into mature seeds (f). For these relationships 92 
correlation tables are necessary. While experience has shown that 
it is desirable that all tables of data should be published, it really 
$ Pearson, K., and others. Phil. Trans. Roy. Soc. London A 197:285-379- 
Igol. 
