Karl Pearson and David Heron 163 
with a bias which may well be called in question has gone out of his way to 
select extreme cases, which had already been indicated by one of us as markedly 
non-Gaussian, but he makes no attempt to measure the wide range of physical 
characters for which the Gaussian is a legitimate practical assumption. Mr Yule 
refers to Dr Macdonell's memoir as a case in which the applicability of the Gaussian 
fourfold table method was " in the first place adequately tested " before adoption *. 
He leaves his uninitiated reader ignorant of two important facts, (i) that in the 
majority of fourfold classifications there is no possibility of such adequate testing 
because only the fourfold division has been provided, and (ii) the test in this case 
was directly made at the suggestion of Pearson and in his Laboratory to test the 
efficiency of the Gaussian method on ordinary data such as form probably nine-tenths 
of the frequencies which occur in practice. The work on this paper of Macdonell's 
began almost immediately on the completion of the theoretical memoir of 1900 on 
the fourfold table, and Mr Yule's statement that the warning of Pearson in the 
fundamental memoir of 1900 that normal correlation was not universal " seems to 
have been forgotten in a few weeks at mostj-" is, as many others of his statements 
from the historical standpoint, hopelessly inaccurate. Thus the paper on eye- 
colour in man and coat-colour in horses was presented in August, 1899, and 
antedates the presentation of the theoretical paper of February, 1900. The 
"warning" could hardly have been promptly forgotten, for the paper was with- 
drawn and rewritten in order to test the value of the method of association then just 
propounded by Mr Yule, and to develop, when that iu as found defective, what, it was 
believed, ivas a better treatment. Mr Yule writes that "Professor Pearson raised no 
objection then and as far as I know has raised no objection since to my coefficient 
Q ; indeed he referred to ' the extreme elegance and simplicity of Mr Yule's 
coefficient of association.' " Naturally when one finds a method wholly inadequate 
one does not turn and rend an old pupil and former colleague. What Pearson did 
do was to test Mr Yule's Q against other similar coefficients and finding it less 
stable than any of them, it was dropped and has never been and never will be 
used in any work done under his supervision. But an interesting point arises 
here, which it is, perhaps, worth mentioning. Endeavouring to find for any 
fourfold division an analogue to Sheppard's median division formula, i.e. for a 
Gaussian fourfold 
a | b 
c | d 
b . 7T a — b 
r -- cos 7r r = sin w r , 
a + b 2 a + b 
Pearson hit upon the fact that the fourfold 
\/bc | 
V 'ad 
* Journ. of R. 8. S. Vol. lxxv. p. 631 
t Ibid. p. 6U. 
