206 
On Criminal Anthropometry 
Now the organs selected by Scotland Yard are very far from being slightly 
correlated. We have seen that Face Breadth and Head Breadth are highly 
correlated, and the correlations between Finger, Cubit, Foot and Height are very 
high. It cannot therefore be said that these seven measurements form an ideal 
group, and I do not suppose they would have been chosen if their correlations had 
been actually known beforehand. My results at this point conflict with those 
of Dr Garson, who writes in his paper in the Journal of the Anthropological 
Institute to which I have already referred : " In a mixed population such as we 
have to deal with in England, the correlation between the different measurements 
used for the classification of criminals is slight." 
Supposing, however, these organs have been selected, we may ask in the next 
place: What is the best order for entering an Index ? This is the problem to which 
we shall next turn, and for its solution I propose to apply a general theorem in 
correlation which may be stated thus : 
(17) Given n variables, then the most probable value A^ of one of the 
variables, for given values J.o, A.^, ... An, of the remaining n—\, is given by the 
equation 
A, 
and its variability Si = \/ , where ctj is the S.D. of the organ ^j, and <7q of the 
organ Aq] the mean of and riiq the mean of Ag-, r^g the correlation 
coefficient of A^ and Ag, and r, 
finally A is the determinant 
the correlation coefficient of Aq and Ag- ; and 
and Rpq the minor corresponding to Vpq. 
When Ar,, A-^, ...An are known for only one individual, the probable error of 
the determination Aj^ is 
■67449 .,^|-; 
when they are known for p individuals, the probable error of A^, which is then the 
corresponding mean value, becomes 
■67449 
pRy. 
When n = 2, the reconstruction formula becomes 
Ai-7ni = — - m2), 
0-2 
