Karl Pearson 
139 
These give : 
Mean: 24-6175. 
/i., = 15-25897, /X, = G02-23008, 
/X, = 19-21542, ;Lt,= 2023-21556, 
leading to the reducing nonic : 
24^9 - 80-8739?' - 13-2924(^^' - 42-3159f/ + 306-5227?^ 
- 166-4257$=' - 24-8654r/' + 12-6008r^ + 1-2081 = 0 
which gives p,, = — \0q = — 7-0031. 
This provides the two components : 
T. minus. T. majus. 
mi = 21-6772, m., = 26-9993, 
a-, = 2-2404, a., = 3-2981, 
«i = 1611-18, Vo = 1988-82. 
The components and their compound are figured in Diagram XIII, p. 140, 
and we find for « = 21, = 52-68 and P = '00016. There has thus been much 
improvement of goodness of fit, although the result is still unsatisfactory. 
It is impossible, however, to look through the graphs given by Sir David Bruce 
and others for the human strains* without being convinced of their fundamentally 
bimodal character, although there appears to be much evidence of its being 
disguised by heterogeneity of host and treatment. 
(viii) Diagram XIV (p. 141) gives the resolution for the human strain from 
Chituluka-f. The constants 
Mean = 26-172, 
/X., = 23-52260, fi, = 1179-30786, 
^, = _ 37-13226, fi, = - 3248-43805, 
leading to the reducing nonic : 
24r/ - 393-8678f/ - 49-6370f/ + 520-29103-^ + 8226-9435(;^ 
- 12493-5620g' - 10ri017f/ + 855-7520g + 63-2383 = 0. 
The value of the root is jOg = — 10(/ = — 16-2295 and this leads to the com- 
ponents given in the Table p. 125, and illustrated in the diagram. The graph 
while giving broadly some of the features of the case is b}' no means a satisfactory 
fit; for ?( = 21 groups, ;j^' = 86 and P is < -000,000,1. The diagram suggests that 
we are probably dealing with a mixture of three components with means about 
18-5, 25-5 and 31-0, but at present we have no satisfactory method of performing- 
multiple resolutions of this character. 
* R. S. Proc. Vol. 86, B, pp. 285—30-2. 
t R. S. Proc. Vol. 86, B, p. 291. 
