W. R. Macdonell 
TABLE 21. 
Probable Errors of S.D. in Criminals. 
203 
By Full Correlation 
Table, 3000 
By Full Correlation 
Table, 1306 
By New 
Method, 3000 
Head Length (cms.) 
Head Breadth (cms.) 
Face Breadth (cm.s.) 
Finger (cms.) 
Height (ins.) 
•0053 
•0044 
•0048 
•0221 
•0079 
•0068 
•0066 
•0072 
•0342 
•0071 
•0101 
•0099 
•0075 
•0495 
F99347 
•77634 
•79048 
1^46909 
•78875 
TABLE 22. 
Probable Errors of r in Criminals. 
By New Method 
Height and Head Breadth (•?■= ^1831) 
Head Length and Face Breadth (r=^3945) 
Head Length and Finger (r=^3007) 
Head Length and Height ()- = ^3399) 
Head Breadth and Cubit (>-=^1352) 
Finger and Cubit (>•= -8464) 
•0210 
•0172 
•0181 
•0207 
•0196 
•0079 
By Formula 
•6745 ( 1-}-^) 
Vnoo 
•0197 
•0172 
•0185 
•0180 
•0200 
•0058 
It will be noted from the results for Head Length and Finger that the 
Probable Errors of the standard deviations by the new method for oOOO are very 
nearly equal to those by the usual method for 1306 when + h-i is large (see § 9), 
that is, when we arrange our 9-fold table so that the middle division is large. A 
reference to § 11 will also show that by arranging our tables in this way we 
obtained remarkably good results for s.D. and mean of Head Length and Finger; 
we might therefore expect improved results if we were to arrange all our 9-fold 
tables throughout on this principle. 
It is to be noted that the Probable Errors of the means by the new method 
are practicall}^ as small as those of the standard deviations. 
Owing to the laborious character of the calculation for finding the probable 
error of r by the new method, I have worked out only eight of the 21 errors 
(see Tables 16 and 22), but these eight being fairly representative will give 
a good idea of the magnitudes involved. 
In the 2nd column of Table 22, the probable error is calculated by the 
usual formula for normal frequency, supposing n = 1100, and on the whole the 
results correspond very closely with those obtained by the new method, except in 
the case of the last coefficient, where the difference is considerable, but as it 
happens the absolute amount of error is small in either case. We may therefore 
18—2 
