The Degree of Accuracy of Statistical Data ii 



where / is the range of the mortality table, here taken to be the 

 60 years, from 25 to 85, he finds : 



^^.718,529,308,595. 



"By a rougher quadrature process for the whole range from 

 20 to 90," he finds : 



/5=:. 80 1,086,783. 



The value of yS, as computed from King and Hardy's c for the 

 range from 17 to 88 years, is: 



/3=r. 804,162,5 



The difference in the values of both /3 and c is due partly to 

 difference of range, partly to method of computation, and hence 

 it is not to be expected that they should agree absolutely. Still, 

 there is no excuse for carrying out the computations to twelve 

 decimals, when dift'erent methods give results which differ in 

 the first figure. Again, the constants can not be more accurate 

 than the data upon which they are based, and no mortality sta- 

 tistics at our disposal are correct even to half the number of 

 places used by Professor Pearson. From trials I have made, I 

 am sure that an eight-place logarithm table would give, on the 

 whole, fully as good results as those obtained by Professor Pear- 

 son, especially as, in spite of the difference in values of the con- 

 stants, the improvement of his results over those obtained by the 

 method of averages "is not very sensible." It is to be deplored 

 that Professor Pearson should mar the effectiveness of his work 

 by his desire for pseudo-accuracy. 



VIII 



In vol. I, part II, of Biometrika, Mr. W. Palin Elderton, an 

 actuary, gives tables for testing curve fitting. These tables, 

 which are six-place tables, have been computed by the aid of 

 eight-, ten-, and eleven-place logarithm and integral tables, in- 

 volving an enormous expenditure of time and energy. Now it 

 so happens in most cases where a large number of observations 

 is involved that these tables are of no value. For instance, in the 



97 



