K. Pearson and A. Lee G3 
11 
Actual 
1 
"696 
•674 
2 
1-176 
1-177 
3 
1 -537 
1-538 
4 
1 -ssa 
1-832 
5 
2-087 
2-086 
6 
2-314 
2-313 
7 
2-529 
2-519 
8 
2-710 
2-710 
9 
2-888 
2-888 
10 
3-055 
3-056 
11 
3-215 
3-216 
12 
3-367 
13 
3-513 
U 
3-654 
15 
3-791 
It appears accordingly probable that our formula up to possibly ?i = 15 or 
more will represent the probable error with not more than "001 error. The 
extreme divergence for the case of ?i = 1 is noteworthy, and seems to place this 
result on a different footing to the others. The general agreement is shown 
graphically in the diagram on p. 64. 
The general use of the table provided will be obvious, it enables us to tell the 
probability of any outlying individual really being a member of a population of 
which the constants are known. Thus one may look forward to the day when 
the biometric constants of a race being sufficiently well known, it may be possible 
to tell from a complex of five or six characters whether a skeleton or a skull may 
be reasonably supposed to have belonged to a member of that race. At present 
the labour of calculating the correlation coefficient-determinants and their minors 
stands in the way of much work in this direction, when we wish to advance 
beyond two or three characters. 
The incomplete normal moment functions provide us with a method of 
determining, theoretically at least, the constants of a truncated normal distribu- 
tion. The method is as follows. Let N be the total frequency, cr the standard 
of 
Mean 
* For plotting the curve we have also the ordinates -3710, -0276, - -0733 for n = }^, 0-1, 0, respectively. 
