M. Greenwood and J. D. C. White 
517 
TABLE VIA. 
Curves Resulting from an Analysis of 15,000 Selected Cells. 
Samples of 25 
Samples of 50 
Samples of 100 
Number of 
Samples 
300 
150 
Mean (in terms 
of Bacilli per 
Sample) 
91-1133 
182-2200 
363-5667 
Mode (in terms 
of Bacilli per 
Sample) 
89-0117 
179-4383 
360-1649 
(7 (m terms 
of Baeilli) 
16-3123 
24-5984 
37-0043 
-0457 
-0980 
•1019 
2- 8211 
3- 7994 
4- 5418 
Material 
m 
/32-3 
Skewness 
P(Fit) 
Samples of 25 
Samples of 50 
Samples of 100 
- -0702 
+ -0582 
+ -0292 
12-4217 
7-4301 
7-2109 
2-7151 
-3-0885 
-1-2890 
6-7230 
5-7783 
- -1789 
+ -7994 
+ 1-5418 
■1288 
-1131 
-0920 
-9966 
-7365 
-2164 
Equations : 
3^=114-724 (1 + ^-3-) (l-j^) . 
Unit of grouping 8 bacilli per 25 cells. 39-5 bacilli per 25 cells taken as zero. 
-7-2109 3 0885 tan- 
6-7230 
Origin at 4:-'12n, where 12 bacilli per 50 cells form the unit of grouping and the zero is at 
105-5. 
r ,.2 -1-4-7151 l-2890tan-i 
2/ = 26-445|^l + j..^t;g3.,J xe 5™3. 
Origin at 5-5642, where 16 bacilli per 100 cells are the unit and tlie zero is at 258-5. 
Next as to the agreement between the theoretical constants of the curves of 
means and their empirical values as above determined. Table VII gives the 
calculated values of and B2 — S [assuming that where n is the number of 
individuals in the sample and fii,-^2 the constants of the original distribution 
/So — 3 
id i?2 — 3=— = 1 toget^Jier with the actual values. 
R = ^ 
n n 
In only one case, that of the 25's from the selected 15,000, is the agreement at 
all possible. " Student," in a very interesting communication*, has suggested that 
the divergence noted by him in our previous curve of 80 means, depended on the 
existence of homotyposis in the samples. This suggestion no doubt contains some 
truth and could be satisfactorily tested by mixing up our individual records of 
leucocytes and drawing really random samples. We have not felt justified in 
delaying publication of our results until we could find leisure to undertake this 
very laborious task, particularly as other factors are involved. This is clear from 
the fact that the divergence is more marked in the case of the larger samples 
which must be less influenced by homotyposis. 
* Biometrika, Vol. vii. p. 210. 
Biometrika vii 66 
