SELECTION ON THE VARIABILITY AND CORRELATION OF ORGANS. 
‘29 
Now this again is a remarkable result; by selecting an organ correlated with 
two others, neither of which are correlated with each other, we have produced a 
considerable correlation, and what is more, one of a negative sign. 
In other words, if humerus and femur were unrelated to each other, but were 
related to stature, then a selection of stature would result in men of long femur 
havino’ a short humerus, and vice versd. 
Illusti'ation IV. —Suppose the correlation between greatest length and breadth 
of the skull to he '25, and between both and the auricular height to be '5. Now let a 
stringent selection, — = of height take place. What modification will there 
be of the length and breadth correlation ? 
Table II. (o), case (j) gives us at once the result— 
tog = ’0033. 
In other words, the correlation between length and breadth would be sensibly 
destroyed by such a selection. Thus correlation can be created or destroyed or 
reversed by selection. 
■ The above illustrations, hypothetical though they may be, will suffice to indicate 
how entirely dependent correlation is upon selection. We must look upon coefficients 
of correlation, in fact, as just as much the outcome of selection as coefficients of 
variation, standard-deviations, or even the mean size of organs. No selection can 
take place, in the sense in which it has usually been understood to take place— i.e., 
by a change of mean and of variability, without at the same time the means, 
variabilities and the correlations of all correlated, but not directly selected, organs 
being varied. This is true whether the non-selected organs be initially correlated or 
not among themselves. We must always bear in mind this all-important funda¬ 
mental conception, that natural or artificial selection, or even random sampling, are 
in themselves active factors in the modification [i.e., creation, destruction, or reversal) 
of correlation. Thus not only is the impossibility of the constancy of correlation 
for local races obvious, but the primary importance of insuring that our samples are 
representative, and not accidentally selected samples, in all observations or experiments 
on heredity, homotyposis, or organic correlation becomes more and more manifest. 
We must not lay too much stress on two heredity constants—differing, for example, by 
more than the probable error of their difference—unless we are convinced, which 
practically it will be difficult to l^e, that all modification of correlation by unin¬ 
tentional and unmarked selection has really been avoided. 
(7.) Let us now take the next most simple case. TfA, B, C, D he four mutually 
correlated organs [in either the same or different individuals), and a selection tahe 
place of A and B, to find the changes in the characters ofi the non-selected 
organs. 
