E. C. Snow 
59 
The notions used to indicate environmental conditions for this new work are 
very simple. Shortly, they consist in making the mortality of a cohort of one sex 
the measure of the environment for the corresponding cohort of the opposite sex 
on which it is desired to ascertain the possible effects of selection. Thus if we 
wish to investigate the selective effect on the male mortality of the third, fourth 
and fifth years of life of variations in that mortality in the first two years, in 
addition to fixing the size of the male cohort we fix the size of the corresponding 
female cohort and also the total female mortality in the first five years of life. We 
can thus suppose that we are dealing with districts in which the female mortalities 
up to five years of age for the cohorts born in a particular year are the same. For 
these districts we find varying male mortalities in each of the periods considered 
(see Table below), and the mean values of these male mortalities in both periods 
throughout the whole series of districts can be found. Do the mortalities in the 
second period of those districts whose mortalities in the first period deviated in 
the positive direction deviate, on the average, in the positive or negative direction? 
Districts with the same female environment will possess varying proportions of 
male weaklings. If these weaklings are killed off in the earlier period, the popula- 
tion which survives to the later one is stronger and likely, therefore, to have a 
smaller mortality, and this would be indicated by a negative correlation between 
the mortalities in the two periods (with the proviso dealt with in § xxxiv of the 
memoir). To the criticism that the total male mortality is highly correlated with 
the total female mortality, and that by making the latter constant we are 
practically fixing the former, we can reply by pointing to the considerable 
standard deviation of the total male mortality when correction is made for 
constant female mortality (see Table below). Evidence of a more general 
character, too, can be gathered by turning over the leaves of any of the Registrar- 
General's valuable Decennial Supplements to his Annual Reports. Pick out a 
few of the registration districts in which the mortality of one sex for any of the 
age-groups given is practically the same and compare the mortalities of the other 
sex among those districts for the same age-group. Quite appreciable variation in 
the numbers will be found to exist. Not many of such districts can actually be 
found, but the method of partial correlation essentially consists of a contrivance 
by which we can for statistical purposes reduce all districts to a constant type. 
No method of measuring environment can be theoretically perfect. Districts 
under the same environmental conditions would not have the same mortality, but 
the latter would be distributed in some way due to random causes. Of two districts 
under the same general environment one may in a particular year suffer to a 
greater extent from epidemics of measles, scarlet fever or summer diarrhoea, and 
part of the problem of selection consists in ascertaining if these epidemics strike 
more at the weaker children than at a random sample of all children, and this is 
ascertained by inquiring if the surviving population is more immune in a subsequent 
period. The two methods which are employed in this paper to measure environ- 
ment are quite distinct and bear no physical relation to each other. In the one 
8-2 
