SELECTION ON THE VAlHAEILTrY AND COEEELATION OF ORGANS. 
G1 
tlie nature of tlie contour lines of the surface of survivals shows tliat the contour lines 
above referred to, and marked “boundary” in the diagram, must touch the Aino 1 in 1000 
limit and the French 1 in 1000 limit respectively at the p(nuts in Avhich they are 
touched by tlie contour lines p — 1 and p' = 1 of the corresponding surfaces of 
survivals. I have already indicated that the major axes of these hoimdaries ai-e 
15'285 for the French aiid 15‘890 for the Aino. The corresponding values of the 
parameter k are respectively given by 
15-285' 
2-7:3^38 
= 5-5911, ^1^ = 7-0578. 
Hence by (cxviii.) Ave can easily find the frequency of population outside the 
ccntnurs Kir and k^; if these be vy and Ave have : 
-000,000,163, vji = -000,000,000,015. 
'flius the French population Avould liave to he extended tt) a boundary in Avhich only 
about 1 in six millions AA^as excluded, and the Aino jiopulation to a boundary excluding 
only 15 in the billion! The boundaries of Avhat Ave may thus term the selection 
populations are far lai-ger than our conventional boundaries of 1 in 1000 for represen¬ 
tative popidations. In fact, it AAmuld be impossible to select a representative Aino 
population from a coiiA^entional representatiAm French jiopidation and vice versa —in 
either case the very excejitional members of French oi- Aino populations are 
required to complete the conventional representatwe populations of Aino or French 
by selection. 
(15.) I have devoted most of my consideration of the surface of survivals to 
a })articular case in Avhich tAvo organs have 1)een selected, and Ave consider the nature 
(.)f p Avhich determines the fraction of each group of individuals Avhich survives. 
1 liaAm done this partly because normal surfaces are at best.only an approximate 
representation of our selectable and selected distributions, and partly because I have 
thought a concrete case Avould best la-ing out the general points of investigations of 
this kind. 
But some little indication of the properties of the surface of survival-rates ought 
to he indicated here, or it may appear that they have been overlooked. While the 
contour lines of the correlation frequency surfaces for two organs must be ellijises, this 
does not folloAv in the case of the surface of survival-rates. In our illustration they 
AA^ei-e ellipses, but they may be also parabolas, hyperbolas, or even straight lines. 
We must not therefore expect to find always a “centre” of positive or negative 
selection. We may come across a “saddle-back system ” of contours Avith the rate 
of survival constant along two intersecting lines, but rising in one pair of opposite 
angles and falling in the other pair. In this case Ave have fields of negative and 
positive selection separated by tAvo independent relations between the two organs, 
