288 
A Statistical Study in Cancer Death- Mates 
death it is the rule to select the diseases ordinarily known as constitutional 
in preference to those known as local diseases. This will be made clear by an 
extract from the New Tables issued by the L.G.B. and the Schedules of causes of 
death : "Thus Cancer should be selected in preference to Pneumonia, and Diabetes 
in preference to Heart-disease." Further the groups already dealt with, viz. 
"Ill-defined and unknown causes" are, perhaps, the best statistical guide to the 
efficiency of the local medical profession. 
Professor Pearson has kindly drawn my attention to another possible source of 
error. If it were shown that both diseases were more likely to attack persons in 
the same social class, or engaged in the same occupations, an artificial correlation 
might then be set up, owing to the inhabitants of the different cities being 
engaged in occupations of a diverse nature. It will be seen later that cancer and 
diabetes are probably negatively correlated in respect to the occupational groups. 
There is I think no doubt that diabetes is a disease especially prone to attack the 
wealthy and sedentary classes, but in the U.S.A. this does not seem to be true of 
cancer. This question is of importance because Mr Heron has by a study of the 
London statistics arrived, by indirect methods, at an opposite conclusion. From 
the U.S.A. Census Reports it was possible to obtain the numbers of men employed 
in various occupations, together with their age distribution, and the cancer and 
diabetes death-rates. These are given first in eight inclusive groups and then 
in sub-groups. For the present purpose correction factors for the death-rates 
were calculated for the eight large groups and for 28 sub-groups as shown in 
the table. The various occupations were then arranged in order according to their 
social status, but as it was unlikely that two people would place them in exactly 
the same order I asked a friend to rearrange them for me and the results of 
both orders are given. The correlation was found by the method of ranks as given 
by Pearson*, = 2 sin -/Ju, where p^., = l - 6S {v^- v^f/N {N- - I). Correlating 
high death-rates with superior social occupations the values are as follows : 
Eight Inclusive Groups. 
Cancer and Occupations ?' = — 'olTG + "IT^G. 
Diabetes and Occupations r = -9374 + -0289. 
28 Suh-gj'oups. 
1st grouping 2nd grouping 
Cancer and Occupations ?• = - -4574 + -1007, r = - -3680 ± -1102. 
Diabetes and Occupations r= -6716 ± '0700, r= -6748 ± -0694. 
To check the results thus found four " four-fold tables " were constructed, the 
division of the death-rates being taken at their mean, and the occupations being 
divided at Nos. 9 and 17 of the order as shown in the table. 
* "On Further Metliods of Measuring Correlation," Drapers' Research Memoirs, Dulau and Co. 
