280 
A Statistical Studi/ in Cancer Death-Rates 
Readers who have fitted curves to " disease " distributions grouped in this 
manner will I think agree, that a fit as good as that given in Fig. 1 indicates the 
homogeneous nature of this group. I have frequently been struck with the 
excellent fit given by diseases, like puerperal septicaemia, where the chance of 
a wrong diagnosis is unlikely, and greatly disappointed with the bad fit obtained 
by a disease that presents difficulty in diagnosis, such for instance as enteric fever 
in hot climates, where the heterogeneous character of the group is undoubted. 
The excellence of fit in the case of the male cancer curve is I think evidence of 
some value as to the homogeneity of these returns. 
The goodness of fit as measured by the usual ')(- formula* for the modified 
male curve is P= "-i and excluding the irregular groups at 2-5 and 107'5, P= 7S. 
The curve in the case of the diabetes distribution is not satisfactory. No 
single curve can be fitted to the figures which will even approximately represent 
the frequency polygon. It seemed advisable to see if two or more curves could be 
found that would serve to express the recorded rates. It is fairly easy to see that 
these curves cannot all be normal. The old age component of diabetes is clearly 
skew, and it was not hard to obtain a good representation of this part of the 
mortality. The earlier mortality appears to consist of two components ; a diabetes 
of youth with a mean age of 15 to 20, and a diabetes of middle age with a mean 
age of about 40. These components would agree to some extent with those found 
by Pearson in his resolution of the general mortality curve f. The discussion of 
this point has been reserved for another paper. Further figures are needful before 
it can be definitely stated that the disease now known as Diabetes melitus can be 
certainly resolved into two or more age groups. The English death returns 
for this disease show the same general features as those obtained from the United 
States j. 
Before any comparisons of the cancer death-rates can be profitably made it is 
necessary that they should be corrected for the age constitution of the population. 
Cancer is mainly a disease of advanced life, and as this is the portion of the 
population most unevenly distributed between town and country, the difference in 
crude death-rates due to this cause must be borne in mind when rates are being 
compared. Similarly no useful comparisons can be made between the rates as 
found for the different occupations until corrected for age. The group of bankers 
for instance has a very different age composition from that of clerks, as the 
following correction factors show: bankers = 0"3516: clerks = 1'3157. Thus the 
crude rates of 414 and 280 become with correction 150 and 368 respectively. 
The usual method § of obtaining the correction factor has been adopted, and 
* Biometrika, Vol. i. p. 155. t Phil. Trans. Vol. 186, p. 407. 
J The female diabetes frequency polygon exhibits the same features as the male distribution, the 
rates being similar. The early age portions of both cancer and diabetes frequencies require special 
consideration, which I hope to give on another occasion. 
§ R — death-rate in standard population, ?•] , 9-2 .. . r„ = death-rates at age periods, .Tj , .Tg . . . .x^ = numbers 
of population in similar age groups, for district to be corrected, where A' is total population. Then 
if li' = .S' (.'■'•)/A', lilll' is the correction factor. 
