Lucy Whitakbr 
60 
and is therefore not a very close approximation, a result shown when we use 
a binomial by the substantial improvement in the measure P of "goodness of 
fit." Even in this case we are not prepared to say what is the population for 
which the q = •01875 in the case of these announcements of deaths of women over 
90 years of age. It can scarcely be that there are only 29 women over 90 years 
TABLE XVI. 
Constants for Deaths of Aged. 
Men. 
Age over 
P 
<1 
Probable 
Error 
of q 
+ -03314 
+ -03349 
+ -02902 
± -02934 
Probable 
Error 
of n 
7)i 
Binomial 
P 
Expo- 
nential 
P 
70 years . . . 
80 years ... 
85 years . . . 
90 years . . . 
1-12965 
1-12152 
1-01903 
1-00654 
- -12965 
- -12152 
- -01903 
- -00654 
-28-8747 
- 14-0703 
-43-2996 
-42-8498 
+ 7 -3734 
± 3-8704 
+ 67-5797 
±192-3069 
3-7436 
1-7099 
-8239 
-2801 
-1355 
•9358 
-9737 
•6741 
-0045 
-1129 
-9715 
-6672 
Women. 
Age over 
V 
2 
Probable 
Error 
of q 
n 
Probable 
Error 
of 11 
m 
Binomial 
P 
Expo- 
nential 
P 
70 years . . . 
80 years . . . 
85 years . . . 
90 years . . . 
1-34012 
1-20770 
1-14507 
•98125 
-- -34012 
- -20770 
- -14507 
+ -01875 
+ -04161 
± -03294 
+ -03077 
+ -02779 
- 10-3522 
-10-4400 
- 8-1447 
+ 29-0573 
+ 1-2307 
+ 1-8309 
+ 1-9627 
+ 43-0634 
3-5210 
2-1569 
1-1816 
•5447 
•8084 
-9686 
-9860 
•9848 
■0000 
-0018 
•1062 
-8116 
of age living in the country, whose deaths are likely to be announced in the Times 
when they occur. Further the probable error of q is such that actually this case 
might equally well be a random sample from material following a negative 
binomial. Analysing our material we see that our first two cases of males and 
the first three of females are such that they could not possibly be random samples 
from positive binomials, the probable errors of q are too small. Next, seven cases 
out of the eight do give actually negative binomials and the eighth might, having 
regard to its probable errors, well be a negative binomial. Tiius although our 
daily occurrences are certainly in Bortkewitsch and Mortara's sense " small numbers," 
they give no support to the use of a Poisson-Exponential. 
If it be said that these " small numbers " differ in character from those used 
by our authors, the reply must be : we know in none of these cases the real 
population from which deaths are to be considered as drawn. The chances of 
death are certainly graduated with age, but the chances of suicide are graduated 
with temperament, and the same is true of alcoholism, or again the chance of 
