322 
On the Fundamental Conceptions of Biology 
When one opponent has not even a preliminary training in biometry, and the other 
fails to attach any clear ideas to the terms used by his antagonist, used apparently 
as if they had universally accepted weight, it seems very hard to find a common 
ground for discussion. Let the reader not suppose this to be an exaggerated 
statement of the case. Consider the terms Variation, Correlation, Regression. 
There is nothing more familiar to the biometrician who has had experience of 
vital statistics than the distinction between a standard deviation measuring 
variability and a coefficient of correlation measuring degree of likeness or 
association. If he has only worked out the constants for one correlation table 
between two different organs he has learnt the distinction between these 
characters. He knows that any degree of correlation may be associated with 
any degree of variability. He knows that regression is not peculiar to heredity nor 
to identity in the organs compared. Now in my memoir I define homotyposis as 
the resemblance of certain like parts, it is therefore a correlation, and whatever its 
numerical value it may be associated as my memoir shows with all sorts of values 
of variation*. This is perfectly obvious to the biometrician so soon as he has 
realised the numerical definitions attached to these terms. Now Mr Bateson 
writes : 
" An ' undifferentiated series of like parts ' means only a series of like parts 
which have varied and are varying among themselves but little. A series of 
highly variable like parts is a series in which differentiation exists or is beginning 
to exist in complex and irregular fashion" {R. S. Proc. Vol. LXIX. p. 197). 
And again : " If differentiation exists and is not recognised the apparent 
homotyposis due to individuality will, as Professor Pearson perceives, be im- 
mediately lowered" (Ibid. p. 169). 
Now I have tried to understand what is the meaning Mr Bateson attaches to 
the terms used in these sentences and it appears to me as a direct result of the 
words cited that high variation is associated with low correlation and vice versa ; 
or that variation and correlation have in Mr Bateson's biological usage a 
significance which is diametrically opposed to their numerical definition by the 
biometrician. We are obviously using the same words for very different quantities. 
Thus our use of the terms variation and correlation is clearly not the same. Nor 
is it better in the matter of regression. Throughout all Mr Bateson's writings, as 
well as in his criticism of my paper, there runs a hopelessly confused notion of 
what we are to understand by regression. The concept of regression is equally 
obscure in Professor Hugo de Vries' ideas on the establishment of breeds. Any 
population tabled for two characters in each individual or in each related pair, 
whether it be a population of coin-tossings, dice-throws, earwigs, or butterflies' 
scales, exhibits the phenomenon of regression, and this whether it is dimorphic 
or monomorphic, or exhibits continuous or discontinuous variation (in one of 
Mr Bateson's senses). All the statistician means by regression is this : If 
* See, for example, p. 327 of my memoir, Phil. Trans., Vol. 197, A., where it is shown how very 
sensibly reducing the variation of a character in the hart's-tongue fern does not sensibly alter correlation. 
