350 
distribution just mentioned. As the equation of this curve 
would vield a surer basis for further inferences, it will 
be of great importance to obtain it. 
Until now, a methodical inquiry into the type of this 
curve has seemed impossible. In practice every investigator, 
having a predilection for a definite type, tries to make 
a curve of that class agree as much as possible with the 
data. À perfect harmony between the analytical curve 
and the observations is never to be expected because of 
the irregularities in the latter. We can even imagine the 
case of two wholly different types both fitting equally 
well. ÂAn absolute criterion, which would enable us to 
decide which is the true type, is wanting. Al we can 
do is to aim at a great probability that we are dealing 
with the true curve. 
The reliability of our conclusions increases in proportion 
as the material is selected more carefully. This should 
be homogeneous, and also numerous: we need a far 
greater number than 300, which is generally considered 
sufficient for a variability coefficient. 
It will be evident that the more a curve agrees with 
the data, the greater the probability of its being the true 
one. Investigators with decided mathematical bent, such 
as Pearson and his followers, are apt to regard the agreement 
as the only criterion. To me it would seem also of 
importance, if we could interpret our curve biologically, 
or in other words, if we could imagine causes which may 
have led to this special type of distribution. 
This, then, is actually the case with Kapteyn's type 
of Skew Frequency Curves, for which I refer to his admi- 
rable mathematical treatment of the relations between 
causes and their effects !). I will cite only some of the 
1) J. C. Kapteyn, Skew Frequency Curves in Biology and Statistics. 
Groningen, Noordhoff. 1903. 
