50 
On the Poisson Lmo of Small Numbers 
I. (b) Suicides of Girls. 
TABLE IV. 
Number of Suicides 
0 
1 
2 
■j' 
Number of Years 
15 
9 
1 
0 
The binomial is : 
25[-7418 + -2582p°« 
Mean --4400 and ^„ = -3264. 
We find g - -2582 + -1012, ?i - 1-7041 ± -7850. 
As in the case of the boys' suicides, if q were practically zero its probable error 
would be + "1908, and there is nothing in this result again to justify us in asserting 
that q is indefinitely small and n indefinitely large. 
Actually we have ; 
TABLE V. 
Number of Suicides per Year. 
0 
1 
2 
3 
Actual 
15 
9 
1 
0 
Bortkewitsch 
16-1 
7-1 
1-8 

Binomial («) 
15-0 
8-9 
1-1 
Binomial (b) 
15-2 
8-7 
ri 
(a) is the binomial considered above, (b) is the binomial obtained by taking 
n a whole number = 2, and q= mean/2 = '22, i.e. 25 ('78 + ■22)-. 
It is clear that either (a) or (b) gives better results than the Poisson-Expo- 
nential. Applying the test of goodness to fit, we have 
= '007 for the binomial (a), 
= '610 for Bortkewitsch's solution. 
Both give P > '60 but the first is much better than the second. 
If both boys and girls are taken together, we find the binomial 
25 (-9333 + •0667)^». 
This is the nearest approach to a small q and big 11 we have so far found — i.e. the 
nearest approach so far to an exponential, but it is reached by a process, i.e. that of 
adding together two series of entirely different means and variabilities in a manner 
which cannot be justified, for Bortkewitsch's hypothesis depends essentially on the 
homogeneity of his material. Even here the fit of the point binomial is slightly 
better than that of the exponential. 
