458 Biometric Workers and Statistical Reviewers 
value of 695 and a standard deviation of 41 it cannot be maintained for a moment that the 
reviewer had any ground whatever for the opinion he stated. (For the reader who has not 
referred to the original memoir I may say that I have found there the magnitude of the 
(negative) correlation which should be expected if there were no selection ; this arising solely 
from the fact that the districts with a high mortality in the first period would have a smaller 
population to be aimed at in the second, and therefore, without selection, a certain amount of 
negative correlation should be expected. But in every case except the Prussian one for the 1881 
cohort — where the (negative) value of 03 r 12 is very high — this correlation is less than the corre- 
sponding probable error of 0 3?'i2-) 
The complaint is also made of my "dangerous measure of environment." I admit that the 
method of measuring environment in the memoir is rather subtle, and that its use requires great 
care. I believe, however, that the real danger lies in the ease with which it may be misunder- 
stood by those who have not the time or inclination to ascertain what it really means. I hope 
soon to complete the development of a criterion which shall give a measure of the goodness of 
any particular correction for environment, etc., but in the meantime I may point out that the 
danger is considerably less than that involved in the correction which necessitates the correlation 
of rates. There is often a likelihood of spurious correlation arising when this is done, and there 
is evidence accumulating that a number of earlier investigations in the field of correlation may 
be invalidated on account of the use of the correlation between rates. The measurement of 
environment has always been a difficulty in statistical researches requiring its use and, for the 
purposes of future work, I shall be grateful if the reviewer can produce any facts invalidating 
the one employed by me, or can make suggestions for a more valuable one. The only improved 
method I have been able to think of is fundamentally similar to the one actually adopted, but 
involves the introduction of more variables. 
I must confess to considerable amazement that the Journal critic should write " It seems to 
vis also that on reconsideration he is not likely to adhere to his present opinion that any in- 
vestigation on the selectivity of the death-rate must be made on a homogeneous group of districts ; 
heterogeneity is really of the essence of the case in the most important form of the problem." 
He can hardly have appreciated the facts I stated on p. 10 and the tables on pp. 19 and 21 of 
the memoir. Briefly, I found that when treating of a mixture of districts I obtained practically 
zero correlations. When I took out from these the rural ones and also the urban ones (there 
was a residue of areas coming completely under neither head) I found very significant negative 
values for the former, but zero correlations for the latter. On this problem and an earlier one 
(Journal, May 1911) I have spent many weeks in analysing heterogeneous material into more 
or less homogeneous groups, and no amount of verbal disquisition as apart from statistical 
analysis is likely to overthrow the experience I gained in those weeks. As I perceive the 
problem of the selectivity of the death-rate, we have statistically to get the populations in a 
large number of districts under uniform environmental conditions. Differences in the mortality 
in a given period will then arise solely from random causes. We then want to investigate to 
what extent the mortality in a district which is below the mean of all districts in that period is 
followed by a mortality which is above the mean of all districts in a following period. The 
difficulty is the statistical one of the measurement of environment. I spent considerable time 
in trying to devise a method of attacking the problem which could be applied to a mixture of 
urban and rural districts, but failed, chiefly on account of the migratory habits of urban 
populations. I claim, then, that the measurement of selection investigated by the method 
I adopted essentially requires the analysis of the material into groups of similar characteristics. 
In the third of the Current Notes dealing with the subject there is a serious misquotation 
from the memoir. I quote the whole of the sentence in which it occurs. "The aims of 
the Report seem fairly evident to any reasonably careful reader, but Mr Snow judges that 
' the investigation was specially intended to discover whether selection exists under modern 
