Karl Pearson and David Heron 
287 
Both results correspond within the limits of the probable errors with the actual 
correlations 78 and '92. These results are far more valid than the association 
results, but of course have not the value of a graph showing the regression line. 
Now here is a case of a man proposing to deal with a most interesting problem, 
for which quite serviceable data exist, led at once from the track of sound treat- 
ment by the application of this fallacious doctrine of association ! And this is far 
from a solitary instance of the harm Mr Yule has done by the publication without 
adequate warning or guidance to his readers of the section of his text-book 
treating' of association. 
Diagram XXV. Association of Natality and Mortality in Parisian Arrondissemeuts. 
120 
110 
100 
■§ 90 
3 80 
o 
I 70 
!" 60 
40 
30 
• 
• • 
• 
1 
• 
11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 
Mortality of Arrondissement. 
(15) Further Criticisms of Mr Yule's Methods of Controversy. 
To certain minor points of Mr Yule's memoir reply will be made in the present 
section. 
(a) Partial Correlations formed from the Normal Coefficient (loc. cit. p. 027). 
Mr Yule is deliberately confusing two different ideas, the correlation of A and B, 
two continuous variables, for a constant value of a third variable C, with the 
correlation of A and B for a given range of values summed under a certain class- 
index or group of class-indices of the variable C. The former is the only sense in 
which the term partial correlation has been hitherto used, and there is no reason 
why Mr Yule should deliberately confuse this sense with a wholly different con- 
ception, that of the correlation of A and B for a sub-universe of ft The whole 
